Update config.json
Browse files- config.json +5 -35
config.json
CHANGED
|
@@ -1,16 +1,10 @@
|
|
| 1 |
{
|
| 2 |
"_name_or_path": "meta-llama/Llama-3.1-8B-Instruct",
|
| 3 |
-
"architectures": [
|
| 4 |
-
"LlamaForCausalLM"
|
| 5 |
-
],
|
| 6 |
"attention_bias": false,
|
| 7 |
"attention_dropout": 0.0,
|
| 8 |
"bos_token_id": 128000,
|
| 9 |
-
"eos_token_id": [
|
| 10 |
-
128001,
|
| 11 |
-
128008,
|
| 12 |
-
128009
|
| 13 |
-
],
|
| 14 |
"hidden_act": "silu",
|
| 15 |
"hidden_size": 4096,
|
| 16 |
"initializer_range": 0.02,
|
|
@@ -28,20 +22,11 @@
|
|
| 28 |
"bnb_4bit_compute_dtype": "bfloat16",
|
| 29 |
"bnb_4bit_quant_storage": "uint8",
|
| 30 |
"bnb_4bit_quant_type": "nf4",
|
| 31 |
-
"bnb_4bit_use_double_quant": true
|
| 32 |
-
"llm_int8_enable_fp32_cpu_offload": false,
|
| 33 |
-
"llm_int8_has_fp16_weight": false,
|
| 34 |
-
"llm_int8_skip_modules": null,
|
| 35 |
-
"llm_int8_threshold": 6.0,
|
| 36 |
-
"load_in_4bit": true,
|
| 37 |
-
"load_in_8bit": false,
|
| 38 |
-
"quant_method": "bitsandbytes"
|
| 39 |
},
|
| 40 |
"rms_norm_eps": 1e-05,
|
| 41 |
"rope_scaling": {
|
| 42 |
"factor": 8.0,
|
| 43 |
-
"high_freq_factor": 4.0,
|
| 44 |
-
"low_freq_factor": 1.0,
|
| 45 |
"original_max_position_embeddings": 8192,
|
| 46 |
"rope_type": "llama3"
|
| 47 |
},
|
|
@@ -53,31 +38,16 @@
|
|
| 53 |
"vocab_size": 128256,
|
| 54 |
"serverless": {
|
| 55 |
"enabled": true,
|
| 56 |
-
"max_batch_size":
|
| 57 |
"max_sequence_length": 2048,
|
| 58 |
"model_parallelism": {
|
| 59 |
"enabled": true,
|
| 60 |
"num_gpus": 1,
|
| 61 |
-
"num_shards":
|
| 62 |
},
|
| 63 |
"quantization": {
|
| 64 |
"enabled": true,
|
| 65 |
"target_dtype": "int8"
|
| 66 |
-
|
| 67 |
}
|
| 68 |
}
|
| 69 |
}
|
| 70 |
-
"serverless": {
|
| 71 |
-
"enabled": true,
|
| 72 |
-
"max_batch_size": 128,
|
| 73 |
-
"max_sequence_length": 2048,
|
| 74 |
-
"model_parallelism": {
|
| 75 |
-
"enabled": true,
|
| 76 |
-
"num_gpus": 1,
|
| 77 |
-
"num_shards": 1
|
| 78 |
-
},
|
| 79 |
-
"quantization": {
|
| 80 |
-
"enabled": true,
|
| 81 |
-
"target_dtype": "int8"
|
| 82 |
-
}
|
| 83 |
-
}
|
|
|
|
| 1 |
{
|
| 2 |
"_name_or_path": "meta-llama/Llama-3.1-8B-Instruct",
|
| 3 |
+
"architectures": ["LlamaForCausalLM"],
|
|
|
|
|
|
|
| 4 |
"attention_bias": false,
|
| 5 |
"attention_dropout": 0.0,
|
| 6 |
"bos_token_id": 128000,
|
| 7 |
+
"eos_token_id": [128001, 128008, 128009],
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
"hidden_act": "silu",
|
| 9 |
"hidden_size": 4096,
|
| 10 |
"initializer_range": 0.02,
|
|
|
|
| 22 |
"bnb_4bit_compute_dtype": "bfloat16",
|
| 23 |
"bnb_4bit_quant_storage": "uint8",
|
| 24 |
"bnb_4bit_quant_type": "nf4",
|
| 25 |
+
"bnb_4bit_use_double_quant": true
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
},
|
| 27 |
"rms_norm_eps": 1e-05,
|
| 28 |
"rope_scaling": {
|
| 29 |
"factor": 8.0,
|
|
|
|
|
|
|
| 30 |
"original_max_position_embeddings": 8192,
|
| 31 |
"rope_type": "llama3"
|
| 32 |
},
|
|
|
|
| 38 |
"vocab_size": 128256,
|
| 39 |
"serverless": {
|
| 40 |
"enabled": true,
|
| 41 |
+
"max_batch_size": 64, // Lowered batch size for better compatibility
|
| 42 |
"max_sequence_length": 2048,
|
| 43 |
"model_parallelism": {
|
| 44 |
"enabled": true,
|
| 45 |
"num_gpus": 1,
|
| 46 |
+
"num_shards": 2 // Adjusted for distributed inference
|
| 47 |
},
|
| 48 |
"quantization": {
|
| 49 |
"enabled": true,
|
| 50 |
"target_dtype": "int8"
|
|
|
|
| 51 |
}
|
| 52 |
}
|
| 53 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|