Synchronizing local compiler cache.

.gitattributes

@@ -3880,3 +3880,7 @@ neuronxcc-2.20.9961.0+0acef03a/MODULE_e828a9ab6bbbe87f2110+a9d440f5/wrapped_neff
 neuronxcc-2.20.9961.0+0acef03a/MODULE_e9ce27504f605841dd3d+ae6a382b/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.20.9961.0+0acef03a/MODULE_f6fe1f7719e8a4b503de+ed72d204/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.20.9961.0+0acef03a/MODULE_ff351d57db88c064c02a+ae6a382b/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.20.9961.0+0acef03a/MODULE_38415f1a88551d65c8cb+ed72d204/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.20.9961.0+0acef03a/MODULE_c839c3566b4b3f8ea4d7+a9d440f5/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.20.9961.0+0acef03a/MODULE_c839c3566b4b3f8ea4d7+a9d440f5/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.20.9961.0+0acef03a/MODULE_cb6febf0237e8ba1d4a2+ae6a382b/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.20.9961.0+0acef03a/0_REGISTRY/0.3.1.dev0/phi3/microsoft/Phi-3-mini-4k-instruct/8c3b00bbcd3136816d3f.json

@@ -0,0 +1,77 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "microsoft/Phi-3-mini-4k-instruct",
+  "_task": "text-generation",
+  "architectures": [
+    "Phi3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "microsoft/Phi-3-mini-4k-instruct--configuration_phi3.Phi3Config",
+    "AutoModelForCausalLM": "microsoft/Phi-3-mini-4k-instruct--modeling_phi3.Phi3ForCausalLM"
+  },
+  "embd_pdrop": 0.0,
+  "hidden_act": "silu",
+  "hidden_size": 3072,
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "max_position_embeddings": 4096,
+  "model_type": "phi3",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "async_mode": false,
+    "attn_kernel_enabled": false,
+    "batch_size": 4,
+    "capacity_factor": null,
+    "cc_pipeline_tiling_factor": 2,
+    "checkpoint_id": "microsoft/Phi-3-mini-4k-instruct",
+    "checkpoint_revision": "0a67737cc96d2554230f90338b163bc6380a2a85",
+    "continuous_batching": true,
+    "enable_bucketing": false,
+    "ep_degree": 1,
+    "flash_decoding_enabled": false,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "is_chunked_prefill": false,
+    "local_ranks_size": 2,
+    "logical_nc_config": 1,
+    "max_batch_size": 4,
+    "max_context_length": 4096,
+    "max_topk": 256,
+    "mlp_kernel_enabled": false,
+    "mlp_kernel_fuse_residual_add": false,
+    "n_active_tokens": 4096,
+    "neuronxcc_version": "2.20.9961.0+0acef03a",
+    "num_cores_per_group": 1,
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.3.1.dev0",
+    "output_logits": false,
+    "padding_side": "right",
+    "pp_degree": 1,
+    "qk_layernorm": false,
+    "qkv_kernel_enabled": false,
+    "rpl_reduce_dtype": "bfloat16",
+    "sequence_length": 4096,
+    "sequence_parallel_enabled": false,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": null,
+    "torch_dtype": "bfloat16",
+    "tp_degree": 2,
+    "vocab_parallel": false
+  },
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 32,
+  "original_max_position_embeddings": 4096,
+  "partial_rotary_factor": 1.0,
+  "resid_pdrop": 0.0,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "sliding_window": 2047,
+  "tie_word_embeddings": false,
+  "use_cache": true,
+  "vocab_size": 32064
+}

neuronxcc-2.20.9961.0+0acef03a/MODULE_38415f1a88551d65c8cb+ed72d204/compile_flags.json

@@ -0,0 +1 @@
+["--target=trn1", "--auto-cast=none", "--model-type=transformer", "--tensorizer-options=--enable-ccop-compute-overlap --cc-pipeline-tiling-factor=2 --vectorize-strided-dma ", "-O2", "--lnc=1", "--logfile=/tmp/nxd_model/context_encoding_model/_tp0_bk0/log-neuron-cc.txt"]

neuronxcc-2.20.9961.0+0acef03a/MODULE_38415f1a88551d65c8cb+ed72d204/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:127de4ebfba08416ddadc2bdabc7cda64274df1428038bdf0bd948e9166f76cd
+size 902324

neuronxcc-2.20.9961.0+0acef03a/MODULE_38415f1a88551d65c8cb+ed72d204/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6036e3d9addc161c09a892a21f5418ecbcbc8515850d2a811b93c66aa41b527
+size 5817344

neuronxcc-2.20.9961.0+0acef03a/MODULE_c839c3566b4b3f8ea4d7+a9d440f5/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0665f8c38436fb626a14d35da9eb597733480d210a0657a26f3caf0a293b1ab8
+size 8827904

neuronxcc-2.20.9961.0+0acef03a/MODULE_c839c3566b4b3f8ea4d7+a9d440f5/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0823904c815ef6554590d443987bf7d260c9f0015a61a007d3aa1647b661beac
+size 8965270

neuronxcc-2.20.9961.0+0acef03a/MODULE_cb6febf0237e8ba1d4a2+ae6a382b/compile_flags.json

@@ -0,0 +1 @@
+"--model-type=transformer -O1 --lnc=1 --internal-hlo2tensorizer-options='--experimental-unsafe-fp8e4m3fn-as-fp8e4m3 --verify-hlo=false' --logfile=/tmp/nxd_model/layout_opt/log-neuron-cc.txt"

neuronxcc-2.20.9961.0+0acef03a/MODULE_cb6febf0237e8ba1d4a2+ae6a382b/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:81c5db0309eddcc2c3313b025fa81c4f09d88c3f03eb3d861f21d6753dab4b9b
+size 136006

neuronxcc-2.20.9961.0+0acef03a/MODULE_cb6febf0237e8ba1d4a2+ae6a382b/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:66cedc854b20b8d1acb60ed5a191d6a9bfe0d3bbdb8c06326b2f2b460fe9f754
+size 3974144