diff --git "a/model.safetensors.index.json" "b/model.safetensors.index.json" --- "a/model.safetensors.index.json" +++ "b/model.safetensors.index.json" @@ -1,8363 +1,899 @@ { "metadata": { "total_parameters": 20914757184, - "total_size": 13440104208 + "total_size": 40884511434 }, "weight_map": { - "lm_head.weight": "model-00003-of-00003.safetensors", - "model.embed_tokens.weight": "model-00001-of-00003.safetensors", - "model.layers.0.input_layernorm.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.0.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.0.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.0.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.0.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.1.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.1.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.1.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.1.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.10.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.10.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.10.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.10.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.11.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.11.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.11.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.11.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.12.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.12.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.12.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.12.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.13.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.13.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.13.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.13.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.14.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.14.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.14.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.14.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.15.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.15.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.15.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.15.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.16.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.16.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.16.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.16.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.17.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.17.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.17.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.17.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.18.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.18.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.18.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.18.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.19.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.19.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.19.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.19.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.2.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.2.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.2.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.2.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.20.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.20.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.20.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.20.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.21.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.21.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.21.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.21.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.22.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.22.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.22.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.22.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.23.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.23.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.23.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.23.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.24.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.24.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.24.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.24.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.25.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.25.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.25.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.25.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.26.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.26.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.26.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.26.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.27.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.27.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.27.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.27.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.28.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.28.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.28.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.28.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.29.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.29.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.29.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.29.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.3.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.3.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.3.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.3.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.30.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.30.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.30.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.30.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.31.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.31.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.31.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.31.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.4.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.4.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.4.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.4.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.5.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.5.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.5.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.5.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.6.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.6.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.6.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.6.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.7.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.7.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.7.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.7.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.8.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.8.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.8.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.8.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.9.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.9.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.9.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.9.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.0.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.0.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.0.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.1.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.1.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.1.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.10.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.10.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.10.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.11.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.11.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.11.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.12.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.12.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.12.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.13.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.13.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.13.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.14.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.14.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.14.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.15.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.15.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.15.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.16.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.16.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.16.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.17.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.17.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.17.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.18.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.18.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.18.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.19.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.19.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.19.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.2.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.2.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.2.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.20.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.20.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.20.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.21.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.21.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.21.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.22.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.22.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.22.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.23.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.23.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.23.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.24.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.24.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.24.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.25.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.25.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.25.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.26.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.26.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.26.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.27.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.27.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.27.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.28.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.28.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.28.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.29.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.29.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.29.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.3.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.3.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.3.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.30.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.30.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.30.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.31.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.31.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.31.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.4.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.4.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.4.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.5.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.5.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.5.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.6.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.6.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.6.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.7.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.7.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.7.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.8.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.8.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.8.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.9.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.9.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.9.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.router.linear.bias": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.router.linear.weight": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.router.linear.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.router.linear.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.k_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.k_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.k_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.o_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.o_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.o_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.q_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.q_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.q_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.sinks": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.v_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.v_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.v_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.0.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.input_layernorm.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.0.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.0.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.0.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.0.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.1.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.1.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.1.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.1.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.10.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.10.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.10.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.10.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.11.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.11.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.11.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.11.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.12.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.12.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.12.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.12.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.13.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.13.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.13.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.13.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.14.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.14.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.14.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.14.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.15.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.15.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.15.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.15.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.16.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.16.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.16.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.16.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.17.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.17.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.17.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.17.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.18.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.18.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.18.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.18.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.19.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.19.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.19.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.19.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.2.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.2.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.2.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.2.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.20.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.20.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.20.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.20.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.21.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.21.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.21.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.21.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.22.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.22.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.22.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.22.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.23.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.23.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.23.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.23.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.24.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.24.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.24.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.24.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.25.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.25.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.25.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.25.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.26.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.26.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.26.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.26.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.27.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.27.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.27.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.27.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.28.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.28.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.28.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.28.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.29.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.29.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.29.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.29.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.3.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.3.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.3.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.3.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.30.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.30.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.30.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.30.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.31.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.31.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.31.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.31.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.4.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.4.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.4.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.4.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.5.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.5.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.5.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.5.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.6.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.6.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.6.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.6.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.7.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.7.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.7.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.7.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.8.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.8.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.8.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.8.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.9.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.9.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.9.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.9.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.0.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.0.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.0.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.1.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.1.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.1.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.10.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.10.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.10.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.11.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.11.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.11.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.12.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.12.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.12.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.13.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.13.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.13.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.14.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.14.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.14.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.15.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.15.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.15.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.16.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.16.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.16.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.17.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.17.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.17.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.18.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.18.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.18.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.19.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.19.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.19.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.2.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.2.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.2.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.20.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.20.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.20.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.21.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.21.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.21.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.22.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.22.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.22.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.23.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.23.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.23.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.24.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.24.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.24.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.25.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.25.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.25.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.26.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.26.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.26.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.27.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.27.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.27.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.28.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.28.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.28.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.29.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.29.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.29.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.3.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.3.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.3.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.30.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.30.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.30.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.31.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.31.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.31.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.4.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.4.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.4.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.5.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.5.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.5.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.6.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.6.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.6.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.7.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.7.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.7.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.8.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.8.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.8.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.9.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.9.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.9.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.router.linear.bias": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.router.linear.weight": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.router.linear.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.router.linear.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.k_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.k_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.k_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.o_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.o_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.o_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.q_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.q_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.q_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.sinks": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.v_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.v_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.v_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.1.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.10.input_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.router.linear.bias": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.router.linear.weight": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.router.linear.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.router.linear.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.post_attention_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.k_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.k_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.o_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.o_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.q_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.q_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.sinks": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.v_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.v_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.10.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.input_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.router.linear.bias": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.router.linear.weight": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.router.linear.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.router.linear.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.post_attention_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.k_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.k_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.o_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.o_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.q_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.q_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.sinks": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.v_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.v_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.11.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.input_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.router.linear.bias": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.router.linear.weight": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.router.linear.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.router.linear.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.post_attention_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.k_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.k_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.o_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.o_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.q_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.q_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.sinks": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.v_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.v_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.12.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.input_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.router.linear.bias": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.router.linear.weight": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.router.linear.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.router.linear.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.post_attention_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.k_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.k_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.o_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.o_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.q_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.q_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.sinks": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.v_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.v_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.13.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.input_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.router.linear.bias": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.router.linear.weight": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.router.linear.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.router.linear.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.post_attention_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.k_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.k_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.o_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.o_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.q_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.q_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.sinks": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.v_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.v_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.14.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.input_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.router.linear.bias": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.router.linear.weight": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.router.linear.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.router.linear.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.post_attention_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.k_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.k_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.o_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.o_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.q_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.q_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.sinks": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.v_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.v_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.15.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.input_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.router.linear.bias": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.router.linear.weight": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.router.linear.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.router.linear.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.post_attention_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.k_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.k_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.o_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.o_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.q_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.q_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.sinks": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.v_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.v_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.16.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.input_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.router.linear.bias": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.router.linear.weight": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.router.linear.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.router.linear.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.post_attention_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.k_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.k_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.o_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.o_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.q_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.q_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.sinks": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.v_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.v_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.17.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.input_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.router.linear.bias": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.router.linear.weight": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.router.linear.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.router.linear.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.post_attention_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.k_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.k_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.o_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.o_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.q_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.q_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.sinks": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.v_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.v_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.18.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.19.input_layernorm.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.0.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.0.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.0.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.0.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.1.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.1.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.1.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.1.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.10.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.10.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.10.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.10.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.11.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.11.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.11.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.11.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.12.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.12.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.12.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.12.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.13.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.13.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.13.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.13.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.14.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.14.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.14.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.14.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.15.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.15.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.15.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.15.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.16.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.16.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.16.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.16.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.17.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.17.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.17.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.17.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.18.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.18.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.18.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.18.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.19.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.19.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.19.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.19.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.2.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.2.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.2.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.2.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.20.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.20.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.20.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.20.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.21.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.21.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.21.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.21.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.22.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.22.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.22.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.22.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.23.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.23.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.23.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.23.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.24.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.24.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.24.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.24.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.25.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.25.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.25.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.25.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.26.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.26.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.26.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.26.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.27.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.27.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.27.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.27.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.28.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.28.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.28.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.28.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.29.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.29.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.29.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.29.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.3.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.3.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.3.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.3.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.30.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.30.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.30.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.30.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.31.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.31.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.31.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.31.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.4.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.4.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.4.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.4.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.5.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.5.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.5.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.5.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.6.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.6.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.6.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.6.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.7.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.7.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.7.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.7.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.8.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.8.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.8.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.8.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.9.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.9.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.9.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.9.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.1.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.1.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.1.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.10.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.10.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.10.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.11.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.11.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.11.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.12.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.12.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.12.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.13.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.13.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.13.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.14.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.14.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.14.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.15.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.15.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.15.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.16.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.16.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.16.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.17.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.17.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.17.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.18.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.18.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.18.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.19.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.19.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.19.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.2.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.2.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.2.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.20.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.20.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.20.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.21.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.21.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.21.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.22.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.22.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.22.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.23.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.23.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.23.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.24.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.24.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.24.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.25.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.25.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.25.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.26.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.26.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.26.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.27.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.27.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.27.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.28.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.28.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.28.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.29.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.29.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.29.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.3.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.3.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.3.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.30.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.30.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.30.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.31.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.31.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.31.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.4.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.4.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.4.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.5.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.5.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.5.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.6.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.6.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.6.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.7.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.7.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.7.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.8.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.8.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.8.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.9.bias": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.9.weight": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.9.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.19.mlp.router.linear.bias": "model-00002-of-00003.safetensors", - "model.layers.19.mlp.router.linear.weight": "model-00002-of-00003.safetensors", - "model.layers.19.mlp.router.linear.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.19.mlp.router.linear.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.19.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.19.post_attention_layernorm.weight": "model-00003-of-00003.safetensors", - "model.layers.19.self_attn.k_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.k_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.o_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.o_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.q_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.q_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.sinks": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.v_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.v_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.19.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.2.input_layernorm.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.0.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.0.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.0.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.0.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.1.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.1.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.1.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.1.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.10.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.10.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.10.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.10.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.11.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.11.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.11.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.11.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.12.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.12.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.12.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.12.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.13.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.13.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.13.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.13.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.14.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.14.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.14.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.14.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.15.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.15.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.15.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.15.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.16.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.16.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.16.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.16.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.17.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.17.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.17.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.17.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.18.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.18.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.18.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.18.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.19.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.19.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.19.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.19.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.2.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.2.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.2.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.2.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.20.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.20.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.20.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.20.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.21.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.21.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.21.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.21.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.22.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.22.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.22.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.22.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.23.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.23.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.23.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.23.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.24.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.24.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.24.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.24.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.25.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.25.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.25.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.25.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.26.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.26.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.26.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.26.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.27.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.27.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.27.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.27.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.28.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.28.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.28.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.28.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.29.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.29.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.29.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.29.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.3.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.3.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.3.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.3.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.30.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.30.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.30.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.30.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.31.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.31.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.31.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.31.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.4.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.4.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.4.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.4.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.5.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.5.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.5.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.5.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.6.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.6.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.6.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.6.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.7.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.7.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.7.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.7.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.8.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.8.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.8.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.8.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.9.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.9.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.9.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.9.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.0.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.0.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.0.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.1.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.1.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.1.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.10.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.10.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.10.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.11.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.11.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.11.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.12.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.12.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.12.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.13.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.13.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.13.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.14.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.14.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.14.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.15.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.15.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.15.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.16.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.16.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.16.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.17.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.17.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.17.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.18.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.18.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.18.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.19.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.19.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.19.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.2.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.2.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.2.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.20.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.20.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.20.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.21.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.21.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.21.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.22.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.22.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.22.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.23.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.23.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.23.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.24.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.24.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.24.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.25.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.25.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.25.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.26.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.26.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.26.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.27.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.27.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.27.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.28.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.28.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.28.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.29.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.29.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.29.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.3.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.3.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.3.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.30.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.30.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.30.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.31.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.31.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.31.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.4.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.4.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.4.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.5.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.5.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.5.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.6.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.6.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.6.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.7.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.7.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.7.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.8.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.8.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.8.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.9.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.9.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.9.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.router.linear.bias": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.router.linear.weight": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.router.linear.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.router.linear.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.k_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.k_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.k_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.o_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.o_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.o_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.q_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.q_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.q_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.q_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.sinks": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.v_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.v_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.v_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.2.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.20.input_layernorm.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.0.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.0.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.0.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.0.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.1.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.1.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.1.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.1.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.10.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.10.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.10.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.10.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.11.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.11.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.11.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.11.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.12.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.12.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.12.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.12.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.13.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.13.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.13.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.13.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.14.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.14.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.14.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.14.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.15.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.15.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.15.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.15.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.16.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.16.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.16.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.16.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.17.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.17.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.17.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.17.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.18.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.18.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.18.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.18.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.19.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.19.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.19.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.19.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.2.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.2.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.2.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.2.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.20.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.20.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.20.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.20.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.21.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.21.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.21.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.21.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.22.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.22.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.22.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.22.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.23.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.23.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.23.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.23.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.24.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.24.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.24.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.24.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.25.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.25.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.25.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.25.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.26.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.26.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.26.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.26.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.27.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.27.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.27.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.27.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.28.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.28.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.28.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.28.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.29.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.29.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.29.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.29.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.3.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.3.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.3.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.3.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.30.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.30.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.30.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.30.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.31.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.31.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.31.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.31.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.4.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.4.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.4.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.4.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.5.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.5.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.5.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.5.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.6.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.6.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.6.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.6.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.7.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.7.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.7.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.7.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.8.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.8.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.8.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.8.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.9.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.9.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.9.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.9.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.0.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.0.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.0.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.1.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.1.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.1.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.10.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.10.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.10.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.11.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.11.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.11.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.12.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.12.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.12.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.13.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.13.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.13.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.14.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.14.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.14.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.15.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.15.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.15.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.16.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.16.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.16.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.17.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.17.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.17.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.18.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.18.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.18.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.19.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.19.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.19.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.2.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.2.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.2.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.20.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.20.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.20.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.21.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.21.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.21.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.22.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.22.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.22.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.23.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.23.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.23.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.24.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.24.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.24.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.25.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.25.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.25.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.26.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.26.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.26.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.27.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.27.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.27.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.28.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.28.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.28.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.29.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.29.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.29.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.3.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.3.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.3.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.30.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.30.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.30.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.31.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.31.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.31.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.4.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.4.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.4.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.5.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.5.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.5.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.6.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.6.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.6.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.7.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.7.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.7.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.8.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.8.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.8.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.9.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.9.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.9.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.router.linear.bias": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.router.linear.weight": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.router.linear.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.router.linear.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.post_attention_layernorm.weight": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.k_proj.bias": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.k_proj.weight": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.k_proj.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.k_proj.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.o_proj.bias": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.o_proj.weight": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.o_proj.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.o_proj.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.q_proj.bias": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.q_proj.weight": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.q_proj.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.q_proj.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.sinks": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.v_proj.bias": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.v_proj.weight": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.v_proj.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.v_proj.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.20.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.input_layernorm.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.0.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.0.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.0.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.0.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.1.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.1.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.1.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.1.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.10.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.10.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.10.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.10.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.11.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.11.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.11.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.11.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.12.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.12.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.12.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.12.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.13.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.13.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.13.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.13.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.14.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.14.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.14.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.14.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.15.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.15.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.15.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.15.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.16.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.16.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.16.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.16.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.17.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.17.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.17.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.17.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.18.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.18.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.18.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.18.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.19.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.19.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.19.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.19.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.2.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.2.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.2.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.2.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.20.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.20.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.20.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.20.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.21.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.21.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.21.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.21.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.22.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.22.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.22.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.22.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.23.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.23.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.23.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.23.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.24.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.24.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.24.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.24.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.25.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.25.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.25.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.25.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.26.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.26.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.26.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.26.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.27.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.27.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.27.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.27.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.28.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.28.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.28.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.28.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.29.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.29.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.29.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.29.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.3.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.3.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.3.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.3.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.30.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.30.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.30.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.30.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.31.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.31.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.31.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.31.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.4.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.4.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.4.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.4.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.5.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.5.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.5.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.5.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.6.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.6.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.6.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.6.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.7.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.7.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.7.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.7.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.8.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.8.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.8.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.8.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.9.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.9.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.9.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.9.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.0.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.0.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.0.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.1.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.1.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.1.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.10.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.10.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.10.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.11.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.11.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.11.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.12.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.12.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.12.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.13.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.13.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.13.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.14.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.14.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.14.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.15.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.15.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.15.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.16.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.16.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.16.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.17.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.17.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.17.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.18.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.18.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.18.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.19.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.19.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.19.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.2.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.2.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.2.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.20.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.20.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.20.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.21.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.21.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.21.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.22.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.22.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.22.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.23.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.23.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.23.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.24.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.24.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.24.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.25.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.25.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.25.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.26.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.26.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.26.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.27.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.27.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.27.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.28.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.28.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.28.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.29.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.29.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.29.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.3.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.3.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.3.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.30.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.30.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.30.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.31.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.31.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.31.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.4.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.4.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.4.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.5.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.5.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.5.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.6.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.6.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.6.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.7.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.7.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.7.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.8.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.8.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.8.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.9.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.9.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.9.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.router.linear.bias": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.router.linear.weight": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.router.linear.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.router.linear.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.post_attention_layernorm.weight": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.k_proj.bias": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.k_proj.weight": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.k_proj.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.k_proj.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.o_proj.bias": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.o_proj.weight": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.o_proj.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.o_proj.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.q_proj.bias": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.q_proj.weight": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.q_proj.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.q_proj.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.sinks": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.v_proj.bias": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.v_proj.weight": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.v_proj.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.v_proj.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.21.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.input_layernorm.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.0.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.0.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.0.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.0.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.1.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.1.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.1.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.1.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.10.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.10.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.10.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.10.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.11.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.11.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.11.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.11.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.12.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.12.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.12.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.12.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.13.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.13.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.13.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.13.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.14.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.14.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.14.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.14.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.15.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.15.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.15.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.15.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.16.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.16.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.16.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.16.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.17.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.17.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.17.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.17.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.18.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.18.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.18.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.18.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.19.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.19.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.19.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.19.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.2.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.2.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.2.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.2.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.20.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.20.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.20.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.20.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.21.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.21.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.21.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.21.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.22.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.22.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.22.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.22.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.23.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.23.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.23.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.23.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.24.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.24.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.24.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.24.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.25.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.25.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.25.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.25.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.26.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.26.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.26.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.26.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.27.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.27.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.27.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.27.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.28.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.28.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.28.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.28.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.29.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.29.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.29.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.29.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.3.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.3.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.3.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.3.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.30.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.30.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.30.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.30.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.31.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.31.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.31.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.31.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.4.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.4.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.4.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.4.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.5.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.5.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.5.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.5.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.6.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.6.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.6.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.6.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.7.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.7.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.7.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.7.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.8.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.8.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.8.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.8.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.9.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.9.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.9.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.9.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.0.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.0.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.0.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.1.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.1.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.1.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.10.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.10.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.10.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.11.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.11.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.11.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.12.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.12.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.12.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.13.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.13.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.13.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.14.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.14.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.14.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.15.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.15.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.15.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.16.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.16.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.16.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.17.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.17.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.17.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.18.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.18.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.18.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.19.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.19.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.19.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.2.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.2.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.2.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.20.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.20.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.20.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.21.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.21.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.21.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.22.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.22.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.22.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.23.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.23.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.23.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.24.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.24.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.24.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.25.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.25.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.25.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.26.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.26.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.26.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.27.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.27.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.27.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.28.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.28.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.28.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.29.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.29.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.29.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.3.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.3.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.3.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.30.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.30.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.30.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.31.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.31.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.31.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.4.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.4.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.4.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.5.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.5.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.5.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.6.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.6.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.6.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.7.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.7.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.7.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.8.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.8.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.8.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.9.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.9.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.9.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.router.linear.bias": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.router.linear.weight": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.router.linear.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.router.linear.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.post_attention_layernorm.weight": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.k_proj.bias": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.k_proj.weight": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.k_proj.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.k_proj.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.o_proj.bias": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.o_proj.weight": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.o_proj.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.o_proj.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.q_proj.bias": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.q_proj.weight": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.q_proj.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.q_proj.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.sinks": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.v_proj.bias": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.v_proj.weight": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.v_proj.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.v_proj.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.22.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.input_layernorm.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.0.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.0.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.0.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.0.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.1.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.1.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.1.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.1.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.10.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.10.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.10.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.10.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.11.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.11.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.11.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.11.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.12.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.12.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.12.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.12.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.13.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.13.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.13.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.13.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.14.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.14.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.14.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.14.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.15.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.15.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.15.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.15.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.16.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.16.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.16.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.16.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.17.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.17.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.17.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.17.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.18.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.18.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.18.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.18.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.19.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.19.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.19.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.19.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.2.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.2.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.2.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.2.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.20.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.20.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.20.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.20.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.21.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.21.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.21.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.21.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.22.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.22.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.22.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.22.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.23.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.23.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.23.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.23.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.24.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.24.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.24.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.24.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.25.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.25.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.25.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.25.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.26.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.26.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.26.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.26.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.27.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.27.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.27.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.27.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.28.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.28.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.28.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.28.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.29.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.29.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.29.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.29.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.3.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.3.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.3.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.3.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.30.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.30.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.30.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.30.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.31.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.31.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.31.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.31.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.4.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.4.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.4.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.4.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.5.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.5.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.5.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.5.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.6.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.6.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.6.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.6.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.7.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.7.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.7.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.7.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.8.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.8.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.8.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.8.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.9.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.9.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.9.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.9.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.0.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.0.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.0.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.1.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.1.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.1.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.10.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.10.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.10.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.11.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.11.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.11.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.12.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.12.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.12.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.13.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.13.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.13.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.14.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.14.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.14.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.15.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.15.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.15.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.16.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.16.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.16.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.17.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.17.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.17.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.18.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.18.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.18.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.19.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.19.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.19.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.2.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.2.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.2.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.20.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.20.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.20.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.21.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.21.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.21.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.22.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.22.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.22.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.23.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.23.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.23.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.24.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.24.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.24.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.25.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.25.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.25.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.26.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.26.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.26.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.27.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.27.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.27.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.28.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.28.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.28.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.29.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.29.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.29.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.3.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.3.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.3.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.30.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.30.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.30.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.31.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.31.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.31.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.4.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.4.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.4.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.5.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.5.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.5.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.6.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.6.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.6.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.7.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.7.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.7.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.8.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.8.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.8.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.9.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.9.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.9.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.router.linear.bias": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.router.linear.weight": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.router.linear.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.router.linear.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.post_attention_layernorm.weight": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.k_proj.bias": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.k_proj.weight": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.k_proj.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.k_proj.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.o_proj.bias": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.o_proj.weight": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.o_proj.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.o_proj.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.q_proj.bias": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.q_proj.weight": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.q_proj.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.q_proj.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.sinks": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.v_proj.bias": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.v_proj.weight": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.v_proj.weight.absmax": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.v_proj.weight.quant_map": "model-00003-of-00003.safetensors", - "model.layers.23.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00003-of-00003.safetensors", - "model.layers.3.input_layernorm.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.0.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.0.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.0.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.0.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.1.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.1.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.1.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.1.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.10.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.10.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.10.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.10.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.11.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.11.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.11.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.11.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.12.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.12.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.12.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.12.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.13.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.13.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.13.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.13.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.14.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.14.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.14.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.14.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.15.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.15.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.15.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.15.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.16.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.16.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.16.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.16.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.17.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.17.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.17.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.17.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.18.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.18.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.18.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.18.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.19.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.19.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.19.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.19.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.2.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.2.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.2.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.2.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.20.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.20.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.20.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.20.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.21.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.21.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.21.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.21.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.22.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.22.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.22.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.22.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.23.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.23.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.23.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.23.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.24.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.24.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.24.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.24.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.25.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.25.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.25.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.25.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.26.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.26.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.26.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.26.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.27.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.27.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.27.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.27.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.28.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.28.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.28.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.28.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.29.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.29.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.29.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.29.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.3.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.3.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.3.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.3.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.30.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.30.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.30.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.30.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.31.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.31.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.31.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.31.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.4.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.4.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.4.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.4.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.5.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.5.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.5.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.5.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.6.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.6.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.6.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.6.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.7.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.7.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.7.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.7.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.8.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.8.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.8.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.8.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.9.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.9.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.9.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.9.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.0.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.0.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.0.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.1.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.1.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.1.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.10.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.10.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.10.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.11.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.11.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.11.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.12.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.12.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.12.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.13.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.13.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.13.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.14.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.14.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.14.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.15.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.15.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.15.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.16.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.16.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.16.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.17.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.17.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.17.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.18.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.18.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.18.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.19.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.19.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.19.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.2.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.2.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.2.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.20.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.20.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.20.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.21.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.21.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.21.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.22.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.22.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.22.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.23.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.23.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.23.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.24.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.24.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.24.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.25.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.25.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.25.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.26.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.26.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.26.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.27.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.27.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.27.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.28.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.28.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.28.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.29.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.29.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.29.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.3.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.3.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.3.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.30.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.30.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.30.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.31.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.31.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.31.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.4.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.4.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.4.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.5.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.5.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.5.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.6.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.6.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.6.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.7.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.7.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.7.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.8.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.8.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.8.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.9.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.9.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.9.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.router.linear.bias": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.router.linear.weight": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.router.linear.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.router.linear.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.k_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.k_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.k_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.k_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.o_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.o_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.o_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.q_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.q_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.q_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.q_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.sinks": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.v_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.v_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.v_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.v_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.3.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.input_layernorm.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.0.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.0.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.0.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.0.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.1.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.1.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.1.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.1.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.10.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.10.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.10.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.10.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.11.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.11.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.11.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.11.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.12.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.12.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.12.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.12.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.13.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.13.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.13.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.13.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.14.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.14.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.14.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.14.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.15.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.15.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.15.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.15.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.16.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.16.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.16.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.16.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.17.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.17.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.17.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.17.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.18.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.18.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.18.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.18.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.19.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.19.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.19.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.19.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.2.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.2.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.2.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.2.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.20.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.20.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.20.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.20.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.21.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.21.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.21.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.21.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.22.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.22.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.22.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.22.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.23.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.23.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.23.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.23.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.24.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.24.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.24.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.24.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.25.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.25.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.25.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.25.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.26.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.26.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.26.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.26.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.27.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.27.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.27.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.27.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.28.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.28.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.28.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.28.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.29.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.29.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.29.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.29.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.3.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.3.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.3.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.3.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.30.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.30.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.30.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.30.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.31.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.31.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.31.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.31.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.4.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.4.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.4.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.4.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.5.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.5.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.5.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.5.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.6.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.6.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.6.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.6.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.7.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.7.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.7.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.7.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.8.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.8.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.8.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.8.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.9.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.9.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.9.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.9.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.0.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.0.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.0.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.1.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.1.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.1.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.10.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.10.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.10.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.11.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.11.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.11.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.12.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.12.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.12.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.13.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.13.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.13.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.14.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.14.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.14.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.15.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.15.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.15.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.16.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.16.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.16.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.17.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.17.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.17.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.18.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.18.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.18.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.19.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.19.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.19.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.2.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.2.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.2.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.20.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.20.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.20.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.21.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.21.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.21.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.22.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.22.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.22.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.23.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.23.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.23.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.24.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.24.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.24.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.25.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.25.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.25.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.26.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.26.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.26.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.27.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.27.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.27.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.28.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.28.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.28.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.29.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.29.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.29.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.3.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.3.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.3.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.30.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.30.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.30.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.31.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.31.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.31.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.4.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.4.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.4.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.5.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.5.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.5.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.6.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.6.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.6.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.7.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.7.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.7.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.8.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.8.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.8.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.9.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.9.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.9.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.router.linear.bias": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.router.linear.weight": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.router.linear.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.router.linear.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.k_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.k_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.k_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.k_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.o_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.o_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.o_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.q_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.q_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.q_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.q_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.sinks": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.v_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.v_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.v_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.v_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.4.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.input_layernorm.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.0.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.0.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.0.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.0.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.1.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.1.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.1.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.1.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.10.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.10.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.10.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.10.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.11.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.11.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.11.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.11.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.12.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.12.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.12.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.12.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.13.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.13.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.13.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.13.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.14.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.14.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.14.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.14.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.15.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.15.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.15.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.15.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.16.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.16.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.16.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.16.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.17.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.17.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.17.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.17.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.18.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.18.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.18.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.18.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.19.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.19.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.19.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.19.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.2.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.2.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.2.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.2.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.20.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.20.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.20.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.20.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.21.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.21.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.21.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.21.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.22.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.22.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.22.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.22.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.23.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.23.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.23.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.23.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.24.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.24.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.24.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.24.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.25.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.25.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.25.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.25.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.26.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.26.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.26.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.26.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.27.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.27.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.27.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.27.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.28.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.28.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.28.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.28.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.29.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.29.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.29.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.29.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.3.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.3.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.3.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.3.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.30.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.30.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.30.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.30.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.31.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.31.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.31.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.31.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.4.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.4.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.4.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.4.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.5.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.5.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.5.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.5.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.6.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.6.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.6.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.6.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.7.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.7.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.7.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.7.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.8.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.8.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.8.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.8.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.9.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.9.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.9.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.9.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.0.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.0.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.0.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.1.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.1.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.1.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.10.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.10.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.10.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.11.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.11.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.11.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.12.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.12.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.12.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.13.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.13.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.13.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.14.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.14.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.14.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.15.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.15.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.15.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.16.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.16.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.16.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.17.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.17.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.17.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.18.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.18.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.18.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.19.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.19.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.19.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.2.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.2.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.2.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.20.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.20.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.20.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.21.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.21.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.21.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.22.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.22.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.22.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.23.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.23.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.23.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.24.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.24.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.24.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.25.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.25.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.25.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.26.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.26.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.26.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.27.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.27.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.27.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.28.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.28.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.28.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.29.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.29.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.29.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.3.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.3.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.3.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.30.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.30.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.30.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.31.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.31.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.31.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.4.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.4.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.4.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.5.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.5.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.5.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.6.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.6.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.6.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.7.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.7.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.7.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.8.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.8.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.8.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.9.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.9.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.9.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.router.linear.bias": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.router.linear.weight": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.router.linear.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.router.linear.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.post_attention_layernorm.weight": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.k_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.k_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.k_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.k_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.o_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.o_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.o_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.o_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.q_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.q_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.q_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.q_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.sinks": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.v_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.v_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.v_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.v_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.5.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.input_layernorm.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.0.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.0.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.0.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.0.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.1.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.1.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.1.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.1.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.10.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.10.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.10.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.10.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.11.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.11.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.11.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.11.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.12.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.12.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.12.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.12.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.13.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.13.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.13.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.13.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.14.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.14.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.14.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.14.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.15.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.15.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.15.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.15.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.16.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.16.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.16.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.16.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.17.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.17.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.17.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.17.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.18.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.18.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.18.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.18.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.19.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.19.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.19.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.19.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.2.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.2.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.2.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.2.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.20.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.20.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.20.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.20.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.21.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.21.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.21.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.21.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.22.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.22.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.22.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.22.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.23.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.23.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.23.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.23.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.24.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.24.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.24.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.24.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.25.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.25.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.25.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.25.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.26.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.26.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.26.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.26.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.27.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.27.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.27.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.27.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.28.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.28.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.28.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.28.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.29.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.29.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.29.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.29.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.3.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.3.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.3.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.3.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.30.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.30.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.30.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.30.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.31.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.31.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.31.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.31.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.4.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.4.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.4.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.4.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.5.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.5.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.5.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.5.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.6.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.6.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.6.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.6.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.7.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.7.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.7.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.7.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.8.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.8.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.8.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.8.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.9.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.9.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.9.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.9.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.0.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.0.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.0.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.1.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.1.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.1.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.10.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.10.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.10.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.11.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.11.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.11.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.12.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.12.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.12.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.13.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.13.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.13.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.14.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.14.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.14.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.15.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.15.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.15.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.16.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.16.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.16.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.17.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.17.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.17.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.18.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.18.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.18.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.19.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.19.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.19.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.2.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.2.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.2.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.20.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.20.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.20.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.21.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.21.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.21.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.22.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.22.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.22.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.23.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.23.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.23.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.24.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.24.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.24.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.25.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.25.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.25.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.26.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.26.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.26.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.27.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.27.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.27.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.28.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.28.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.28.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.29.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.29.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.29.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.3.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.3.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.3.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.30.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.30.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.30.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.31.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.31.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.31.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.4.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.4.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.4.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.5.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.5.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.5.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.6.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.6.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.6.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.7.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.7.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.7.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.8.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.8.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.8.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.9.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.9.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.9.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.router.linear.bias": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.router.linear.weight": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.router.linear.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.router.linear.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.post_attention_layernorm.weight": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.k_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.k_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.k_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.k_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.o_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.o_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.o_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.o_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.q_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.q_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.q_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.q_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.sinks": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.v_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.v_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.v_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.v_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.6.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.input_layernorm.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.0.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.0.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.0.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.0.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.1.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.1.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.1.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.1.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.10.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.10.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.10.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.10.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.11.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.11.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.11.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.11.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.12.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.12.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.12.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.12.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.13.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.13.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.13.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.13.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.14.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.14.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.14.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.14.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.15.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.15.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.15.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.15.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.16.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.16.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.16.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.16.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.17.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.17.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.17.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.17.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.18.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.18.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.18.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.18.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.19.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.19.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.19.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.19.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.2.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.2.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.2.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.2.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.20.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.20.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.20.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.20.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.21.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.21.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.21.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.21.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.22.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.22.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.22.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.22.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.23.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.23.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.23.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.23.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.24.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.24.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.24.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.24.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.25.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.25.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.25.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.25.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.26.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.26.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.26.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.26.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.27.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.27.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.27.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.27.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.28.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.28.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.28.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.28.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.29.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.29.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.29.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.29.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.3.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.3.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.3.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.3.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.30.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.30.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.30.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.30.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.31.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.31.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.31.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.31.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.4.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.4.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.4.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.4.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.5.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.5.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.5.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.5.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.6.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.6.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.6.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.6.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.7.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.7.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.7.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.7.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.8.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.8.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.8.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.8.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.9.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.9.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.9.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.9.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.0.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.0.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.0.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.1.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.1.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.1.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.10.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.10.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.10.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.11.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.11.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.11.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.12.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.12.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.12.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.13.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.13.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.13.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.14.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.14.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.14.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.15.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.15.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.15.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.16.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.16.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.16.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.17.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.17.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.17.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.18.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.18.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.18.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.19.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.19.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.19.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.2.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.2.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.2.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.20.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.20.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.20.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.21.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.21.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.21.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.22.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.22.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.22.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.23.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.23.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.23.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.24.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.24.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.24.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.25.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.25.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.25.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.26.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.26.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.26.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.27.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.27.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.27.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.28.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.28.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.28.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.29.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.29.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.29.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.3.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.3.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.3.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.30.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.30.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.30.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.31.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.31.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.31.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.4.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.4.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.4.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.5.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.5.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.5.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.6.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.6.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.6.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.7.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.7.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.7.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.8.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.8.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.8.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.9.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.9.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.9.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.router.linear.bias": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.router.linear.weight": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.router.linear.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.router.linear.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.post_attention_layernorm.weight": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.k_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.k_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.k_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.k_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.o_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.o_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.o_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.o_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.q_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.q_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.q_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.q_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.sinks": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.v_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.v_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.v_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.v_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.7.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.8.input_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.0.bias": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.0.weight": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.0.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.1.bias": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.1.weight": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.1.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.10.bias": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.10.weight": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.10.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.11.bias": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.11.weight": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.11.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.2.bias": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.2.weight": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.2.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.3.bias": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.3.weight": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.3.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.4.bias": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.4.weight": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.4.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.5.bias": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.5.weight": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.5.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.6.bias": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.6.weight": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.6.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.7.bias": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.7.weight": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.7.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.8.bias": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.8.weight": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.8.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.9.bias": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.9.weight": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.9.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.router.linear.bias": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.router.linear.weight": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.router.linear.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.router.linear.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.8.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.8.post_attention_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.8.self_attn.k_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.k_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.k_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.k_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.o_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.o_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.o_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.o_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.q_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.q_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.q_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.q_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.sinks": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.v_proj.bias": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.v_proj.weight": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.v_proj.weight.absmax": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.v_proj.weight.quant_map": "model-00001-of-00003.safetensors", - "model.layers.8.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00001-of-00003.safetensors", - "model.layers.9.input_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.down_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.0.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.0.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.0.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.0.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.0.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.1.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.1.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.1.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.1.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.1.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.10.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.10.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.10.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.10.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.10.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.11.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.11.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.11.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.11.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.11.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.12.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.12.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.12.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.12.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.12.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.13.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.13.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.13.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.13.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.13.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.14.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.14.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.14.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.14.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.14.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.15.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.15.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.15.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.15.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.15.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.16.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.16.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.16.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.16.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.16.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.17.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.17.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.17.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.17.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.17.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.18.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.18.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.18.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.18.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.18.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.19.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.19.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.19.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.19.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.19.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.2.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.2.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.2.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.2.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.2.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.20.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.20.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.20.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.20.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.20.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.21.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.21.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.21.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.21.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.21.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.22.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.22.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.22.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.22.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.22.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.23.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.23.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.23.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.23.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.23.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.24.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.24.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.24.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.24.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.24.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.25.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.25.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.25.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.25.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.25.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.26.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.26.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.26.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.26.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.26.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.27.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.27.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.27.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.27.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.27.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.28.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.28.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.28.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.28.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.28.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.29.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.29.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.29.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.29.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.29.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.3.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.3.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.3.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.3.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.3.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.30.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.30.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.30.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.30.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.30.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.31.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.31.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.31.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.31.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.31.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.4.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.4.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.4.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.4.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.4.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.5.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.5.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.5.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.5.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.5.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.6.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.6.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.6.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.6.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.6.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.7.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.7.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.7.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.7.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.7.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.8.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.8.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.8.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.8.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.8.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.9.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.9.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.9.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.9.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.experts.gate_up_projs.9.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.router.linear.bias": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.router.linear.weight": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.router.linear.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.router.linear.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.mlp.router.linear.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.post_attention_layernorm.weight": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.k_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.k_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.k_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.o_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.o_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.o_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.q_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.q_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.q_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.sinks": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.v_proj.bias": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.v_proj.weight": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors", - "model.layers.9.self_attn.v_proj.weight.quant_state.bitsandbytes__fp4": "model-00002-of-00003.safetensors", - "model.norm.weight": "model-00003-of-00003.safetensors" + "lm_head.weight": "model-00009-of-00009.safetensors", + "model.embed_tokens.weight": "model-00001-of-00009.safetensors", + "model.layers.0.input_layernorm.weight": "model-00001-of-00009.safetensors", + "model.layers.0.mlp.experts.down_proj": "model-00001-of-00009.safetensors", + "model.layers.0.mlp.experts.down_proj_bias": "model-00001-of-00009.safetensors", + "model.layers.0.mlp.experts.gate_up_proj": "model-00001-of-00009.safetensors", + "model.layers.0.mlp.experts.gate_up_proj_bias": "model-00001-of-00009.safetensors", + "model.layers.0.mlp.router.bias": "model-00001-of-00009.safetensors", + "model.layers.0.mlp.router.weight": "model-00001-of-00009.safetensors", + "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.k_proj.bias": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.k_proj.weight.absmax": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.k_proj.weight.nested_absmax": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.k_proj.weight.nested_quant_map": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.k_proj.weight.quant_map": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.o_proj.bias": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.o_proj.weight.absmax": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.o_proj.weight.nested_absmax": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.o_proj.weight.nested_quant_map": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.o_proj.weight.quant_map": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.q_proj.bias": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.q_proj.weight.absmax": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.q_proj.weight.nested_absmax": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.q_proj.weight.nested_quant_map": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.q_proj.weight.quant_map": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.sinks": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.v_proj.bias": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.v_proj.weight.absmax": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.v_proj.weight.nested_absmax": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.v_proj.weight.nested_quant_map": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.v_proj.weight.quant_map": "model-00001-of-00009.safetensors", + "model.layers.0.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00009.safetensors", + "model.layers.1.input_layernorm.weight": "model-00001-of-00009.safetensors", + "model.layers.1.mlp.experts.down_proj": "model-00001-of-00009.safetensors", + "model.layers.1.mlp.experts.down_proj_bias": "model-00001-of-00009.safetensors", + "model.layers.1.mlp.experts.gate_up_proj": "model-00001-of-00009.safetensors", + "model.layers.1.mlp.experts.gate_up_proj_bias": "model-00001-of-00009.safetensors", + "model.layers.1.mlp.router.bias": "model-00001-of-00009.safetensors", + "model.layers.1.mlp.router.weight": "model-00001-of-00009.safetensors", + "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.k_proj.bias": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.k_proj.weight.absmax": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.k_proj.weight.nested_absmax": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.k_proj.weight.nested_quant_map": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.k_proj.weight.quant_map": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.o_proj.bias": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.o_proj.weight.absmax": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.o_proj.weight.nested_absmax": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.o_proj.weight.nested_quant_map": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.o_proj.weight.quant_map": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.q_proj.bias": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.q_proj.weight.absmax": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.q_proj.weight.nested_absmax": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.q_proj.weight.nested_quant_map": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.q_proj.weight.quant_map": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.sinks": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.v_proj.bias": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.v_proj.weight.absmax": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.v_proj.weight.nested_absmax": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.v_proj.weight.nested_quant_map": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.v_proj.weight.quant_map": "model-00001-of-00009.safetensors", + "model.layers.1.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00009.safetensors", + "model.layers.10.input_layernorm.weight": "model-00004-of-00009.safetensors", + "model.layers.10.mlp.experts.down_proj": "model-00004-of-00009.safetensors", + "model.layers.10.mlp.experts.down_proj_bias": "model-00004-of-00009.safetensors", + "model.layers.10.mlp.experts.gate_up_proj": "model-00004-of-00009.safetensors", + "model.layers.10.mlp.experts.gate_up_proj_bias": "model-00004-of-00009.safetensors", + "model.layers.10.mlp.router.bias": "model-00004-of-00009.safetensors", + "model.layers.10.mlp.router.weight": "model-00004-of-00009.safetensors", + "model.layers.10.post_attention_layernorm.weight": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.k_proj.bias": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.k_proj.weight": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.k_proj.weight.absmax": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.k_proj.weight.nested_absmax": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.k_proj.weight.nested_quant_map": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.k_proj.weight.quant_map": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.o_proj.bias": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.o_proj.weight": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.o_proj.weight.absmax": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.o_proj.weight.nested_absmax": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.o_proj.weight.nested_quant_map": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.o_proj.weight.quant_map": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.q_proj.bias": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.q_proj.weight": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.q_proj.weight.absmax": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.q_proj.weight.nested_absmax": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.q_proj.weight.nested_quant_map": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.q_proj.weight.quant_map": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.sinks": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.v_proj.bias": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.v_proj.weight": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.v_proj.weight.absmax": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.v_proj.weight.nested_absmax": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.v_proj.weight.nested_quant_map": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.v_proj.weight.quant_map": "model-00004-of-00009.safetensors", + "model.layers.10.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00004-of-00009.safetensors", + "model.layers.11.input_layernorm.weight": "model-00005-of-00009.safetensors", + "model.layers.11.mlp.experts.down_proj": "model-00005-of-00009.safetensors", + "model.layers.11.mlp.experts.down_proj_bias": "model-00005-of-00009.safetensors", + "model.layers.11.mlp.experts.gate_up_proj": "model-00005-of-00009.safetensors", + "model.layers.11.mlp.experts.gate_up_proj_bias": "model-00005-of-00009.safetensors", + "model.layers.11.mlp.router.bias": "model-00004-of-00009.safetensors", + "model.layers.11.mlp.router.weight": "model-00004-of-00009.safetensors", + "model.layers.11.post_attention_layernorm.weight": "model-00005-of-00009.safetensors", + "model.layers.11.self_attn.k_proj.bias": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.k_proj.weight": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.k_proj.weight.absmax": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.k_proj.weight.nested_absmax": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.k_proj.weight.nested_quant_map": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.k_proj.weight.quant_map": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.o_proj.bias": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.o_proj.weight": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.o_proj.weight.absmax": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.o_proj.weight.nested_absmax": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.o_proj.weight.nested_quant_map": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.o_proj.weight.quant_map": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.q_proj.bias": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.q_proj.weight": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.q_proj.weight.absmax": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.q_proj.weight.nested_absmax": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.q_proj.weight.nested_quant_map": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.q_proj.weight.quant_map": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.sinks": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.v_proj.bias": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.v_proj.weight": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.v_proj.weight.absmax": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.v_proj.weight.nested_absmax": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.v_proj.weight.nested_quant_map": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.v_proj.weight.quant_map": "model-00004-of-00009.safetensors", + "model.layers.11.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00004-of-00009.safetensors", + "model.layers.12.input_layernorm.weight": "model-00005-of-00009.safetensors", + "model.layers.12.mlp.experts.down_proj": "model-00005-of-00009.safetensors", + "model.layers.12.mlp.experts.down_proj_bias": "model-00005-of-00009.safetensors", + "model.layers.12.mlp.experts.gate_up_proj": "model-00005-of-00009.safetensors", + "model.layers.12.mlp.experts.gate_up_proj_bias": "model-00005-of-00009.safetensors", + "model.layers.12.mlp.router.bias": "model-00005-of-00009.safetensors", + "model.layers.12.mlp.router.weight": "model-00005-of-00009.safetensors", + "model.layers.12.post_attention_layernorm.weight": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.k_proj.bias": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.k_proj.weight": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.k_proj.weight.absmax": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.k_proj.weight.nested_absmax": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.k_proj.weight.nested_quant_map": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.k_proj.weight.quant_map": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.o_proj.bias": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.o_proj.weight": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.o_proj.weight.absmax": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.o_proj.weight.nested_absmax": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.o_proj.weight.nested_quant_map": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.o_proj.weight.quant_map": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.q_proj.bias": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.q_proj.weight": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.q_proj.weight.absmax": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.q_proj.weight.nested_absmax": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.q_proj.weight.nested_quant_map": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.q_proj.weight.quant_map": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.sinks": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.v_proj.bias": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.v_proj.weight": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.v_proj.weight.absmax": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.v_proj.weight.nested_absmax": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.v_proj.weight.nested_quant_map": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.v_proj.weight.quant_map": "model-00005-of-00009.safetensors", + "model.layers.12.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00005-of-00009.safetensors", + "model.layers.13.input_layernorm.weight": "model-00005-of-00009.safetensors", + "model.layers.13.mlp.experts.down_proj": "model-00005-of-00009.safetensors", + "model.layers.13.mlp.experts.down_proj_bias": "model-00005-of-00009.safetensors", + "model.layers.13.mlp.experts.gate_up_proj": "model-00005-of-00009.safetensors", + "model.layers.13.mlp.experts.gate_up_proj_bias": "model-00005-of-00009.safetensors", + "model.layers.13.mlp.router.bias": "model-00005-of-00009.safetensors", + "model.layers.13.mlp.router.weight": "model-00005-of-00009.safetensors", + "model.layers.13.post_attention_layernorm.weight": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.k_proj.bias": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.k_proj.weight": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.k_proj.weight.absmax": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.k_proj.weight.nested_absmax": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.k_proj.weight.nested_quant_map": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.k_proj.weight.quant_map": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.o_proj.bias": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.o_proj.weight": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.o_proj.weight.absmax": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.o_proj.weight.nested_absmax": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.o_proj.weight.nested_quant_map": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.o_proj.weight.quant_map": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.q_proj.bias": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.q_proj.weight": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.q_proj.weight.absmax": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.q_proj.weight.nested_absmax": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.q_proj.weight.nested_quant_map": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.q_proj.weight.quant_map": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.sinks": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.v_proj.bias": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.v_proj.weight": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.v_proj.weight.absmax": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.v_proj.weight.nested_absmax": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.v_proj.weight.nested_quant_map": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.v_proj.weight.quant_map": "model-00005-of-00009.safetensors", + "model.layers.13.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00005-of-00009.safetensors", + "model.layers.14.input_layernorm.weight": "model-00006-of-00009.safetensors", + "model.layers.14.mlp.experts.down_proj": "model-00006-of-00009.safetensors", + "model.layers.14.mlp.experts.down_proj_bias": "model-00006-of-00009.safetensors", + "model.layers.14.mlp.experts.gate_up_proj": "model-00006-of-00009.safetensors", + "model.layers.14.mlp.experts.gate_up_proj_bias": "model-00006-of-00009.safetensors", + "model.layers.14.mlp.router.bias": "model-00005-of-00009.safetensors", + "model.layers.14.mlp.router.weight": "model-00005-of-00009.safetensors", + "model.layers.14.post_attention_layernorm.weight": "model-00006-of-00009.safetensors", + "model.layers.14.self_attn.k_proj.bias": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.k_proj.weight": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.k_proj.weight.absmax": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.k_proj.weight.nested_absmax": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.k_proj.weight.nested_quant_map": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.k_proj.weight.quant_map": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.o_proj.bias": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.o_proj.weight": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.o_proj.weight.absmax": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.o_proj.weight.nested_absmax": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.o_proj.weight.nested_quant_map": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.o_proj.weight.quant_map": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.q_proj.bias": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.q_proj.weight": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.q_proj.weight.absmax": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.q_proj.weight.nested_absmax": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.q_proj.weight.nested_quant_map": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.q_proj.weight.quant_map": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.sinks": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.v_proj.bias": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.v_proj.weight": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.v_proj.weight.absmax": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.v_proj.weight.nested_absmax": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.v_proj.weight.nested_quant_map": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.v_proj.weight.quant_map": "model-00005-of-00009.safetensors", + "model.layers.14.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00005-of-00009.safetensors", + "model.layers.15.input_layernorm.weight": "model-00006-of-00009.safetensors", + "model.layers.15.mlp.experts.down_proj": "model-00006-of-00009.safetensors", + "model.layers.15.mlp.experts.down_proj_bias": "model-00006-of-00009.safetensors", + "model.layers.15.mlp.experts.gate_up_proj": "model-00006-of-00009.safetensors", + "model.layers.15.mlp.experts.gate_up_proj_bias": "model-00006-of-00009.safetensors", + "model.layers.15.mlp.router.bias": "model-00006-of-00009.safetensors", + "model.layers.15.mlp.router.weight": "model-00006-of-00009.safetensors", + "model.layers.15.post_attention_layernorm.weight": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.k_proj.bias": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.k_proj.weight": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.k_proj.weight.absmax": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.k_proj.weight.nested_absmax": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.k_proj.weight.nested_quant_map": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.k_proj.weight.quant_map": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.o_proj.bias": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.o_proj.weight": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.o_proj.weight.absmax": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.o_proj.weight.nested_absmax": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.o_proj.weight.nested_quant_map": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.o_proj.weight.quant_map": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.q_proj.bias": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.q_proj.weight": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.q_proj.weight.absmax": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.q_proj.weight.nested_absmax": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.q_proj.weight.nested_quant_map": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.q_proj.weight.quant_map": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.sinks": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.v_proj.bias": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.v_proj.weight": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.v_proj.weight.absmax": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.v_proj.weight.nested_absmax": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.v_proj.weight.nested_quant_map": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.v_proj.weight.quant_map": "model-00006-of-00009.safetensors", + "model.layers.15.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00006-of-00009.safetensors", + "model.layers.16.input_layernorm.weight": "model-00006-of-00009.safetensors", + "model.layers.16.mlp.experts.down_proj": "model-00006-of-00009.safetensors", + "model.layers.16.mlp.experts.down_proj_bias": "model-00006-of-00009.safetensors", + "model.layers.16.mlp.experts.gate_up_proj": "model-00006-of-00009.safetensors", + "model.layers.16.mlp.experts.gate_up_proj_bias": "model-00006-of-00009.safetensors", + "model.layers.16.mlp.router.bias": "model-00006-of-00009.safetensors", + "model.layers.16.mlp.router.weight": "model-00006-of-00009.safetensors", + "model.layers.16.post_attention_layernorm.weight": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.k_proj.bias": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.k_proj.weight": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.k_proj.weight.absmax": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.k_proj.weight.nested_absmax": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.k_proj.weight.nested_quant_map": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.k_proj.weight.quant_map": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.o_proj.bias": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.o_proj.weight": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.o_proj.weight.absmax": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.o_proj.weight.nested_absmax": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.o_proj.weight.nested_quant_map": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.o_proj.weight.quant_map": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.q_proj.bias": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.q_proj.weight": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.q_proj.weight.absmax": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.q_proj.weight.nested_absmax": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.q_proj.weight.nested_quant_map": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.q_proj.weight.quant_map": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.sinks": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.v_proj.bias": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.v_proj.weight": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.v_proj.weight.absmax": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.v_proj.weight.nested_absmax": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.v_proj.weight.nested_quant_map": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.v_proj.weight.quant_map": "model-00006-of-00009.safetensors", + "model.layers.16.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00006-of-00009.safetensors", + "model.layers.17.input_layernorm.weight": "model-00007-of-00009.safetensors", + "model.layers.17.mlp.experts.down_proj": "model-00007-of-00009.safetensors", + "model.layers.17.mlp.experts.down_proj_bias": "model-00007-of-00009.safetensors", + "model.layers.17.mlp.experts.gate_up_proj": "model-00007-of-00009.safetensors", + "model.layers.17.mlp.experts.gate_up_proj_bias": "model-00007-of-00009.safetensors", + "model.layers.17.mlp.router.bias": "model-00006-of-00009.safetensors", + "model.layers.17.mlp.router.weight": "model-00006-of-00009.safetensors", + "model.layers.17.post_attention_layernorm.weight": "model-00007-of-00009.safetensors", + "model.layers.17.self_attn.k_proj.bias": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.k_proj.weight": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.k_proj.weight.absmax": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.k_proj.weight.nested_absmax": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.k_proj.weight.nested_quant_map": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.k_proj.weight.quant_map": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.o_proj.bias": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.o_proj.weight": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.o_proj.weight.absmax": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.o_proj.weight.nested_absmax": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.o_proj.weight.nested_quant_map": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.o_proj.weight.quant_map": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.q_proj.bias": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.q_proj.weight": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.q_proj.weight.absmax": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.q_proj.weight.nested_absmax": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.q_proj.weight.nested_quant_map": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.q_proj.weight.quant_map": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.sinks": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.v_proj.bias": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.v_proj.weight": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.v_proj.weight.absmax": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.v_proj.weight.nested_absmax": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.v_proj.weight.nested_quant_map": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.v_proj.weight.quant_map": "model-00006-of-00009.safetensors", + "model.layers.17.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00006-of-00009.safetensors", + "model.layers.18.input_layernorm.weight": "model-00007-of-00009.safetensors", + "model.layers.18.mlp.experts.down_proj": "model-00007-of-00009.safetensors", + "model.layers.18.mlp.experts.down_proj_bias": "model-00007-of-00009.safetensors", + "model.layers.18.mlp.experts.gate_up_proj": "model-00007-of-00009.safetensors", + "model.layers.18.mlp.experts.gate_up_proj_bias": "model-00007-of-00009.safetensors", + "model.layers.18.mlp.router.bias": "model-00007-of-00009.safetensors", + "model.layers.18.mlp.router.weight": "model-00007-of-00009.safetensors", + "model.layers.18.post_attention_layernorm.weight": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.k_proj.bias": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.k_proj.weight": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.k_proj.weight.absmax": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.k_proj.weight.nested_absmax": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.k_proj.weight.nested_quant_map": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.k_proj.weight.quant_map": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.o_proj.bias": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.o_proj.weight": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.o_proj.weight.absmax": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.o_proj.weight.nested_absmax": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.o_proj.weight.nested_quant_map": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.o_proj.weight.quant_map": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.q_proj.bias": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.q_proj.weight": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.q_proj.weight.absmax": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.q_proj.weight.nested_absmax": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.q_proj.weight.nested_quant_map": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.q_proj.weight.quant_map": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.sinks": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.v_proj.bias": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.v_proj.weight": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.v_proj.weight.absmax": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.v_proj.weight.nested_absmax": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.v_proj.weight.nested_quant_map": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.v_proj.weight.quant_map": "model-00007-of-00009.safetensors", + "model.layers.18.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00007-of-00009.safetensors", + "model.layers.19.input_layernorm.weight": "model-00007-of-00009.safetensors", + "model.layers.19.mlp.experts.down_proj": "model-00007-of-00009.safetensors", + "model.layers.19.mlp.experts.down_proj_bias": "model-00007-of-00009.safetensors", + "model.layers.19.mlp.experts.gate_up_proj": "model-00007-of-00009.safetensors", + "model.layers.19.mlp.experts.gate_up_proj_bias": "model-00007-of-00009.safetensors", + "model.layers.19.mlp.router.bias": "model-00007-of-00009.safetensors", + "model.layers.19.mlp.router.weight": "model-00007-of-00009.safetensors", + "model.layers.19.post_attention_layernorm.weight": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.k_proj.bias": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.k_proj.weight": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.k_proj.weight.absmax": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.k_proj.weight.nested_absmax": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.k_proj.weight.nested_quant_map": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.k_proj.weight.quant_map": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.o_proj.bias": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.o_proj.weight": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.o_proj.weight.absmax": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.o_proj.weight.nested_absmax": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.o_proj.weight.nested_quant_map": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.o_proj.weight.quant_map": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.q_proj.bias": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.q_proj.weight": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.q_proj.weight.absmax": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.q_proj.weight.nested_absmax": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.q_proj.weight.nested_quant_map": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.q_proj.weight.quant_map": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.sinks": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.v_proj.bias": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.v_proj.weight": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.v_proj.weight.absmax": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.v_proj.weight.nested_absmax": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.v_proj.weight.nested_quant_map": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.v_proj.weight.quant_map": "model-00007-of-00009.safetensors", + "model.layers.19.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00007-of-00009.safetensors", + "model.layers.2.input_layernorm.weight": "model-00002-of-00009.safetensors", + "model.layers.2.mlp.experts.down_proj": "model-00002-of-00009.safetensors", + "model.layers.2.mlp.experts.down_proj_bias": "model-00002-of-00009.safetensors", + "model.layers.2.mlp.experts.gate_up_proj": "model-00002-of-00009.safetensors", + "model.layers.2.mlp.experts.gate_up_proj_bias": "model-00002-of-00009.safetensors", + "model.layers.2.mlp.router.bias": "model-00001-of-00009.safetensors", + "model.layers.2.mlp.router.weight": "model-00001-of-00009.safetensors", + "model.layers.2.post_attention_layernorm.weight": "model-00002-of-00009.safetensors", + "model.layers.2.self_attn.k_proj.bias": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.k_proj.weight.absmax": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.k_proj.weight.nested_absmax": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.k_proj.weight.nested_quant_map": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.k_proj.weight.quant_map": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.o_proj.bias": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.o_proj.weight.absmax": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.o_proj.weight.nested_absmax": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.o_proj.weight.nested_quant_map": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.o_proj.weight.quant_map": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.q_proj.bias": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.q_proj.weight": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.q_proj.weight.absmax": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.q_proj.weight.nested_absmax": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.q_proj.weight.nested_quant_map": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.q_proj.weight.quant_map": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.sinks": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.v_proj.bias": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.v_proj.weight.absmax": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.v_proj.weight.nested_absmax": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.v_proj.weight.nested_quant_map": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.v_proj.weight.quant_map": "model-00001-of-00009.safetensors", + "model.layers.2.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00009.safetensors", + "model.layers.20.input_layernorm.weight": "model-00008-of-00009.safetensors", + "model.layers.20.mlp.experts.down_proj": "model-00008-of-00009.safetensors", + "model.layers.20.mlp.experts.down_proj_bias": "model-00008-of-00009.safetensors", + "model.layers.20.mlp.experts.gate_up_proj": "model-00008-of-00009.safetensors", + "model.layers.20.mlp.experts.gate_up_proj_bias": "model-00008-of-00009.safetensors", + "model.layers.20.mlp.router.bias": "model-00007-of-00009.safetensors", + "model.layers.20.mlp.router.weight": "model-00007-of-00009.safetensors", + "model.layers.20.post_attention_layernorm.weight": "model-00008-of-00009.safetensors", + "model.layers.20.self_attn.k_proj.bias": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.k_proj.weight": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.k_proj.weight.absmax": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.k_proj.weight.nested_absmax": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.k_proj.weight.nested_quant_map": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.k_proj.weight.quant_map": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.o_proj.bias": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.o_proj.weight": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.o_proj.weight.absmax": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.o_proj.weight.nested_absmax": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.o_proj.weight.nested_quant_map": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.o_proj.weight.quant_map": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.q_proj.bias": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.q_proj.weight": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.q_proj.weight.absmax": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.q_proj.weight.nested_absmax": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.q_proj.weight.nested_quant_map": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.q_proj.weight.quant_map": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.sinks": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.v_proj.bias": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.v_proj.weight": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.v_proj.weight.absmax": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.v_proj.weight.nested_absmax": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.v_proj.weight.nested_quant_map": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.v_proj.weight.quant_map": "model-00007-of-00009.safetensors", + "model.layers.20.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00007-of-00009.safetensors", + "model.layers.21.input_layernorm.weight": "model-00008-of-00009.safetensors", + "model.layers.21.mlp.experts.down_proj": "model-00008-of-00009.safetensors", + "model.layers.21.mlp.experts.down_proj_bias": "model-00008-of-00009.safetensors", + "model.layers.21.mlp.experts.gate_up_proj": "model-00008-of-00009.safetensors", + "model.layers.21.mlp.experts.gate_up_proj_bias": "model-00008-of-00009.safetensors", + "model.layers.21.mlp.router.bias": "model-00008-of-00009.safetensors", + "model.layers.21.mlp.router.weight": "model-00008-of-00009.safetensors", + "model.layers.21.post_attention_layernorm.weight": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.k_proj.bias": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.k_proj.weight": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.k_proj.weight.absmax": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.k_proj.weight.nested_absmax": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.k_proj.weight.nested_quant_map": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.k_proj.weight.quant_map": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.o_proj.bias": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.o_proj.weight": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.o_proj.weight.absmax": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.o_proj.weight.nested_absmax": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.o_proj.weight.nested_quant_map": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.o_proj.weight.quant_map": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.q_proj.bias": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.q_proj.weight": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.q_proj.weight.absmax": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.q_proj.weight.nested_absmax": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.q_proj.weight.nested_quant_map": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.q_proj.weight.quant_map": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.sinks": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.v_proj.bias": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.v_proj.weight": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.v_proj.weight.absmax": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.v_proj.weight.nested_absmax": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.v_proj.weight.nested_quant_map": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.v_proj.weight.quant_map": "model-00008-of-00009.safetensors", + "model.layers.21.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00008-of-00009.safetensors", + "model.layers.22.input_layernorm.weight": "model-00008-of-00009.safetensors", + "model.layers.22.mlp.experts.down_proj": "model-00008-of-00009.safetensors", + "model.layers.22.mlp.experts.down_proj_bias": "model-00008-of-00009.safetensors", + "model.layers.22.mlp.experts.gate_up_proj": "model-00008-of-00009.safetensors", + "model.layers.22.mlp.experts.gate_up_proj_bias": "model-00008-of-00009.safetensors", + "model.layers.22.mlp.router.bias": "model-00008-of-00009.safetensors", + "model.layers.22.mlp.router.weight": "model-00008-of-00009.safetensors", + "model.layers.22.post_attention_layernorm.weight": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.k_proj.bias": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.k_proj.weight": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.k_proj.weight.absmax": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.k_proj.weight.nested_absmax": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.k_proj.weight.nested_quant_map": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.k_proj.weight.quant_map": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.o_proj.bias": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.o_proj.weight": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.o_proj.weight.absmax": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.o_proj.weight.nested_absmax": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.o_proj.weight.nested_quant_map": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.o_proj.weight.quant_map": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.q_proj.bias": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.q_proj.weight": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.q_proj.weight.absmax": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.q_proj.weight.nested_absmax": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.q_proj.weight.nested_quant_map": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.q_proj.weight.quant_map": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.sinks": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.v_proj.bias": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.v_proj.weight": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.v_proj.weight.absmax": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.v_proj.weight.nested_absmax": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.v_proj.weight.nested_quant_map": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.v_proj.weight.quant_map": "model-00008-of-00009.safetensors", + "model.layers.22.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00008-of-00009.safetensors", + "model.layers.23.input_layernorm.weight": "model-00009-of-00009.safetensors", + "model.layers.23.mlp.experts.down_proj": "model-00009-of-00009.safetensors", + "model.layers.23.mlp.experts.down_proj_bias": "model-00009-of-00009.safetensors", + "model.layers.23.mlp.experts.gate_up_proj": "model-00009-of-00009.safetensors", + "model.layers.23.mlp.experts.gate_up_proj_bias": "model-00009-of-00009.safetensors", + "model.layers.23.mlp.router.bias": "model-00008-of-00009.safetensors", + "model.layers.23.mlp.router.weight": "model-00008-of-00009.safetensors", + "model.layers.23.post_attention_layernorm.weight": "model-00009-of-00009.safetensors", + "model.layers.23.self_attn.k_proj.bias": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.k_proj.weight": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.k_proj.weight.absmax": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.k_proj.weight.nested_absmax": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.k_proj.weight.nested_quant_map": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.k_proj.weight.quant_map": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.o_proj.bias": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.o_proj.weight": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.o_proj.weight.absmax": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.o_proj.weight.nested_absmax": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.o_proj.weight.nested_quant_map": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.o_proj.weight.quant_map": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.q_proj.bias": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.q_proj.weight": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.q_proj.weight.absmax": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.q_proj.weight.nested_absmax": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.q_proj.weight.nested_quant_map": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.q_proj.weight.quant_map": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.sinks": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.v_proj.bias": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.v_proj.weight": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.v_proj.weight.absmax": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.v_proj.weight.nested_absmax": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.v_proj.weight.nested_quant_map": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.v_proj.weight.quant_map": "model-00008-of-00009.safetensors", + "model.layers.23.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00008-of-00009.safetensors", + "model.layers.3.input_layernorm.weight": "model-00002-of-00009.safetensors", + "model.layers.3.mlp.experts.down_proj": "model-00002-of-00009.safetensors", + "model.layers.3.mlp.experts.down_proj_bias": "model-00002-of-00009.safetensors", + "model.layers.3.mlp.experts.gate_up_proj": "model-00002-of-00009.safetensors", + "model.layers.3.mlp.experts.gate_up_proj_bias": "model-00002-of-00009.safetensors", + "model.layers.3.mlp.router.bias": "model-00002-of-00009.safetensors", + "model.layers.3.mlp.router.weight": "model-00002-of-00009.safetensors", + "model.layers.3.post_attention_layernorm.weight": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.k_proj.bias": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.k_proj.weight": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.k_proj.weight.absmax": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.k_proj.weight.nested_absmax": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.k_proj.weight.nested_quant_map": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.k_proj.weight.quant_map": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.o_proj.bias": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.o_proj.weight": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.o_proj.weight.absmax": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.o_proj.weight.nested_absmax": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.o_proj.weight.nested_quant_map": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.o_proj.weight.quant_map": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.q_proj.bias": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.q_proj.weight": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.q_proj.weight.absmax": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.q_proj.weight.nested_absmax": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.q_proj.weight.nested_quant_map": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.q_proj.weight.quant_map": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.sinks": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.v_proj.bias": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.v_proj.weight": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.v_proj.weight.absmax": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.v_proj.weight.nested_absmax": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.v_proj.weight.nested_quant_map": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.v_proj.weight.quant_map": "model-00002-of-00009.safetensors", + "model.layers.3.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00009.safetensors", + "model.layers.4.input_layernorm.weight": "model-00002-of-00009.safetensors", + "model.layers.4.mlp.experts.down_proj": "model-00002-of-00009.safetensors", + "model.layers.4.mlp.experts.down_proj_bias": "model-00002-of-00009.safetensors", + "model.layers.4.mlp.experts.gate_up_proj": "model-00002-of-00009.safetensors", + "model.layers.4.mlp.experts.gate_up_proj_bias": "model-00002-of-00009.safetensors", + "model.layers.4.mlp.router.bias": "model-00002-of-00009.safetensors", + "model.layers.4.mlp.router.weight": "model-00002-of-00009.safetensors", + "model.layers.4.post_attention_layernorm.weight": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.k_proj.bias": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.k_proj.weight": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.k_proj.weight.absmax": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.k_proj.weight.nested_absmax": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.k_proj.weight.nested_quant_map": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.k_proj.weight.quant_map": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.o_proj.bias": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.o_proj.weight": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.o_proj.weight.absmax": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.o_proj.weight.nested_absmax": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.o_proj.weight.nested_quant_map": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.o_proj.weight.quant_map": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.q_proj.bias": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.q_proj.weight": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.q_proj.weight.absmax": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.q_proj.weight.nested_absmax": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.q_proj.weight.nested_quant_map": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.q_proj.weight.quant_map": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.sinks": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.v_proj.bias": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.v_proj.weight": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.v_proj.weight.absmax": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.v_proj.weight.nested_absmax": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.v_proj.weight.nested_quant_map": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.v_proj.weight.quant_map": "model-00002-of-00009.safetensors", + "model.layers.4.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00009.safetensors", + "model.layers.5.input_layernorm.weight": "model-00003-of-00009.safetensors", + "model.layers.5.mlp.experts.down_proj": "model-00003-of-00009.safetensors", + "model.layers.5.mlp.experts.down_proj_bias": "model-00003-of-00009.safetensors", + "model.layers.5.mlp.experts.gate_up_proj": "model-00003-of-00009.safetensors", + "model.layers.5.mlp.experts.gate_up_proj_bias": "model-00003-of-00009.safetensors", + "model.layers.5.mlp.router.bias": "model-00002-of-00009.safetensors", + "model.layers.5.mlp.router.weight": "model-00002-of-00009.safetensors", + "model.layers.5.post_attention_layernorm.weight": "model-00003-of-00009.safetensors", + "model.layers.5.self_attn.k_proj.bias": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.k_proj.weight": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.k_proj.weight.absmax": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.k_proj.weight.nested_absmax": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.k_proj.weight.nested_quant_map": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.k_proj.weight.quant_map": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.o_proj.bias": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.o_proj.weight": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.o_proj.weight.absmax": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.o_proj.weight.nested_absmax": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.o_proj.weight.nested_quant_map": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.o_proj.weight.quant_map": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.q_proj.bias": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.q_proj.weight": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.q_proj.weight.absmax": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.q_proj.weight.nested_absmax": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.q_proj.weight.nested_quant_map": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.q_proj.weight.quant_map": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.sinks": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.v_proj.bias": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.v_proj.weight": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.v_proj.weight.absmax": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.v_proj.weight.nested_absmax": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.v_proj.weight.nested_quant_map": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.v_proj.weight.quant_map": "model-00002-of-00009.safetensors", + "model.layers.5.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00009.safetensors", + "model.layers.6.input_layernorm.weight": "model-00003-of-00009.safetensors", + "model.layers.6.mlp.experts.down_proj": "model-00003-of-00009.safetensors", + "model.layers.6.mlp.experts.down_proj_bias": "model-00003-of-00009.safetensors", + "model.layers.6.mlp.experts.gate_up_proj": "model-00003-of-00009.safetensors", + "model.layers.6.mlp.experts.gate_up_proj_bias": "model-00003-of-00009.safetensors", + "model.layers.6.mlp.router.bias": "model-00003-of-00009.safetensors", + "model.layers.6.mlp.router.weight": "model-00003-of-00009.safetensors", + "model.layers.6.post_attention_layernorm.weight": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.k_proj.bias": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.k_proj.weight": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.k_proj.weight.absmax": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.k_proj.weight.nested_absmax": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.k_proj.weight.nested_quant_map": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.k_proj.weight.quant_map": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.o_proj.bias": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.o_proj.weight": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.o_proj.weight.absmax": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.o_proj.weight.nested_absmax": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.o_proj.weight.nested_quant_map": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.o_proj.weight.quant_map": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.q_proj.bias": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.q_proj.weight": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.q_proj.weight.absmax": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.q_proj.weight.nested_absmax": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.q_proj.weight.nested_quant_map": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.q_proj.weight.quant_map": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.sinks": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.v_proj.bias": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.v_proj.weight": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.v_proj.weight.absmax": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.v_proj.weight.nested_absmax": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.v_proj.weight.nested_quant_map": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.v_proj.weight.quant_map": "model-00003-of-00009.safetensors", + "model.layers.6.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00009.safetensors", + "model.layers.7.input_layernorm.weight": "model-00003-of-00009.safetensors", + "model.layers.7.mlp.experts.down_proj": "model-00003-of-00009.safetensors", + "model.layers.7.mlp.experts.down_proj_bias": "model-00003-of-00009.safetensors", + "model.layers.7.mlp.experts.gate_up_proj": "model-00003-of-00009.safetensors", + "model.layers.7.mlp.experts.gate_up_proj_bias": "model-00003-of-00009.safetensors", + "model.layers.7.mlp.router.bias": "model-00003-of-00009.safetensors", + "model.layers.7.mlp.router.weight": "model-00003-of-00009.safetensors", + "model.layers.7.post_attention_layernorm.weight": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.k_proj.bias": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.k_proj.weight": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.k_proj.weight.absmax": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.k_proj.weight.nested_absmax": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.k_proj.weight.nested_quant_map": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.k_proj.weight.quant_map": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.o_proj.bias": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.o_proj.weight": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.o_proj.weight.absmax": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.o_proj.weight.nested_absmax": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.o_proj.weight.nested_quant_map": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.o_proj.weight.quant_map": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.q_proj.bias": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.q_proj.weight": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.q_proj.weight.absmax": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.q_proj.weight.nested_absmax": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.q_proj.weight.nested_quant_map": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.q_proj.weight.quant_map": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.sinks": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.v_proj.bias": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.v_proj.weight": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.v_proj.weight.absmax": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.v_proj.weight.nested_absmax": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.v_proj.weight.nested_quant_map": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.v_proj.weight.quant_map": "model-00003-of-00009.safetensors", + "model.layers.7.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00009.safetensors", + "model.layers.8.input_layernorm.weight": "model-00004-of-00009.safetensors", + "model.layers.8.mlp.experts.down_proj": "model-00004-of-00009.safetensors", + "model.layers.8.mlp.experts.down_proj_bias": "model-00004-of-00009.safetensors", + "model.layers.8.mlp.experts.gate_up_proj": "model-00004-of-00009.safetensors", + "model.layers.8.mlp.experts.gate_up_proj_bias": "model-00004-of-00009.safetensors", + "model.layers.8.mlp.router.bias": "model-00003-of-00009.safetensors", + "model.layers.8.mlp.router.weight": "model-00003-of-00009.safetensors", + "model.layers.8.post_attention_layernorm.weight": "model-00004-of-00009.safetensors", + "model.layers.8.self_attn.k_proj.bias": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.k_proj.weight": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.k_proj.weight.absmax": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.k_proj.weight.nested_absmax": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.k_proj.weight.nested_quant_map": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.k_proj.weight.quant_map": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.o_proj.bias": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.o_proj.weight": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.o_proj.weight.absmax": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.o_proj.weight.nested_absmax": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.o_proj.weight.nested_quant_map": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.o_proj.weight.quant_map": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.q_proj.bias": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.q_proj.weight": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.q_proj.weight.absmax": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.q_proj.weight.nested_absmax": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.q_proj.weight.nested_quant_map": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.q_proj.weight.quant_map": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.sinks": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.v_proj.bias": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.v_proj.weight": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.v_proj.weight.absmax": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.v_proj.weight.nested_absmax": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.v_proj.weight.nested_quant_map": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.v_proj.weight.quant_map": "model-00003-of-00009.safetensors", + "model.layers.8.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00009.safetensors", + "model.layers.9.input_layernorm.weight": "model-00004-of-00009.safetensors", + "model.layers.9.mlp.experts.down_proj": "model-00004-of-00009.safetensors", + "model.layers.9.mlp.experts.down_proj_bias": "model-00004-of-00009.safetensors", + "model.layers.9.mlp.experts.gate_up_proj": "model-00004-of-00009.safetensors", + "model.layers.9.mlp.experts.gate_up_proj_bias": "model-00004-of-00009.safetensors", + "model.layers.9.mlp.router.bias": "model-00004-of-00009.safetensors", + "model.layers.9.mlp.router.weight": "model-00004-of-00009.safetensors", + "model.layers.9.post_attention_layernorm.weight": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.k_proj.bias": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.k_proj.weight": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.k_proj.weight.absmax": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.k_proj.weight.nested_absmax": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.k_proj.weight.nested_quant_map": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.k_proj.weight.quant_map": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.o_proj.bias": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.o_proj.weight": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.o_proj.weight.absmax": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.o_proj.weight.nested_absmax": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.o_proj.weight.nested_quant_map": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.o_proj.weight.quant_map": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.q_proj.bias": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.q_proj.weight": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.q_proj.weight.absmax": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.q_proj.weight.nested_absmax": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.q_proj.weight.nested_quant_map": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.q_proj.weight.quant_map": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.sinks": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.v_proj.bias": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.v_proj.weight": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.v_proj.weight.absmax": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.v_proj.weight.nested_absmax": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.v_proj.weight.nested_quant_map": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.v_proj.weight.quant_map": "model-00004-of-00009.safetensors", + "model.layers.9.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00004-of-00009.safetensors", + "model.norm.weight": "model-00009-of-00009.safetensors" } }