Synchronizing local compiler cache.

.gitattributes

@@ -17801,3 +17801,6 @@ neuronxcc-2.26.6360.0+6f180f47/MODULE_cfc98d6eaa0cefa6f7ca+84337dd9/wrapped_neff
 neuronxcc-2.26.6360.0+6f180f47/MODULE_3a4f0acd9d5104d5fff2+8f3c2f7c/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.26.6360.0+6f180f47/MODULE_3a4f0acd9d5104d5fff2+8f3c2f7c/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.26.6360.0+6f180f47/MODULE_84423616ad3ab8d098f4+26701361/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.26.6360.0+6f180f47/MODULE_40bd35cba048a4a8313c+12dd9119/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.26.6360.0+6f180f47/MODULE_97cb9711da676912271f+84337dd9/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.26.6360.0+6f180f47/MODULE_97cb9711da676912271f+84337dd9/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.26.6360.0+6f180f47/0_REGISTRY/0.4.7.dev0/c54cbc73d9a74e547bf7ca1feb2b290b641ed261e32f7c08baba5633884f1298/158491b4d3d885902a4e.json

@@ -0,0 +1,63 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "meta-llama/Meta-Llama-3.1-8B",
+  "_task": "text-generation",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "dtype": "bfloat16",
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 14336,
+  "max_position_embeddings": 131072,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 1,
+    "capacity_factor": null,
+    "checkpoint_id": "meta-llama/Meta-Llama-3.1-8B",
+    "checkpoint_revision": "d04e592bb4f6aa9cfee91e2e20afa771667e1d4b",
+    "continuous_batching": false,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 1,
+    "max_context_length": 4096,
+    "max_topk": 256,
+    "n_active_tokens": 4096,
+    "neuronxcc_version": "2.26.6360.0+6f180f47",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.7.dev0",
+    "output_logits": false,
+    "pp_degree": 1,
+    "prefill_chunk_size": 1024,
+    "sequence_length": 4096,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": "trn1",
+    "torch_dtype": "bfloat16",
+    "tp_degree": 2
+  },
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 8,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": {
+    "factor": 8.0,
+    "high_freq_factor": 4.0,
+    "low_freq_factor": 1.0,
+    "original_max_position_embeddings": 8192,
+    "rope_type": "llama3"
+  },
+  "rope_theta": 500000.0,
+  "tie_word_embeddings": false,
+  "use_cache": true,
+  "vocab_size": 128256
+}

neuronxcc-2.26.6360.0+6f180f47/0_REGISTRY/0.4.7.dev0/llama/meta-llama/Meta-Llama-3.1-8B/158491b4d3d885902a4e.json

@@ -0,0 +1,63 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "meta-llama/Meta-Llama-3.1-8B",
+  "_task": "text-generation",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "dtype": "bfloat16",
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 14336,
+  "max_position_embeddings": 131072,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 1,
+    "capacity_factor": null,
+    "checkpoint_id": "meta-llama/Meta-Llama-3.1-8B",
+    "checkpoint_revision": "d04e592bb4f6aa9cfee91e2e20afa771667e1d4b",
+    "continuous_batching": false,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 1,
+    "max_context_length": 4096,
+    "max_topk": 256,
+    "n_active_tokens": 4096,
+    "neuronxcc_version": "2.26.6360.0+6f180f47",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.7.dev0",
+    "output_logits": false,
+    "pp_degree": 1,
+    "prefill_chunk_size": 1024,
+    "sequence_length": 4096,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": "trn1",
+    "torch_dtype": "bfloat16",
+    "tp_degree": 2
+  },
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 8,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": {
+    "factor": 8.0,
+    "high_freq_factor": 4.0,
+    "low_freq_factor": 1.0,
+    "original_max_position_embeddings": 8192,
+    "rope_type": "llama3"
+  },
+  "rope_theta": 500000.0,
+  "tie_word_embeddings": false,
+  "use_cache": true,
+  "vocab_size": 128256
+}

neuronxcc-2.26.6360.0+6f180f47/MODULE_40bd35cba048a4a8313c+12dd9119/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", "--verbose=35", "--logfile=/tmp/nxd_model/chunked_prefill/_tp0_bk0/log-neuron-cc.txt"]

neuronxcc-2.26.6360.0+6f180f47/MODULE_40bd35cba048a4a8313c+12dd9119/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:228026f9f000b871ed6678ecfe1c079787332799f149e3d6a4734244c65f8ad3
+size 4331401

neuronxcc-2.26.6360.0+6f180f47/MODULE_40bd35cba048a4a8313c+12dd9119/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43f1d5fb01db254aa74e0098748b68062633ce66aec118fef57bbb24eb025553
+size 5110784

neuronxcc-2.26.6360.0+6f180f47/MODULE_97cb9711da676912271f+84337dd9/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", "--verbose=35", "--logfile=/tmp/nxd_model/token_generation/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]

neuronxcc-2.26.6360.0+6f180f47/MODULE_97cb9711da676912271f+84337dd9/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f4aa76dc0b646a4cafd514bd23ac3396849ed218dbbed67e66eaa2f8dcd3d6c
+size 3314685

neuronxcc-2.26.6360.0+6f180f47/MODULE_97cb9711da676912271f+84337dd9/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:661b3872e6f86b7e4c0174cdeb3d760c0f274c2648d5bbdc03ce6bc309e371d8
+size 8920064

neuronxcc-2.26.6360.0+6f180f47/MODULE_97cb9711da676912271f+84337dd9/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8926dcbb545c32197c84c8e471fb12a6f2c41002bfc3aa77980ebfd6f8178657
+size 9067994