Synchronizing local compiler cache.

.gitattributes

@@ -11900,3 +11900,6 @@ neuronxcc-2.21.18209.0+043b1bf7/MODULE_b85364df3fc31288dc1c+d0d57c8a/wrapped_nef
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_0eacf9caad4198400cf5+747527b0/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_0eacf9caad4198400cf5+747527b0/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_fdd85f4d4b3ddd2e04dd+877608f3/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_226601715d0474f1ab25+877608f3/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_fbaa2e54172cefbeabb2+747527b0/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_fbaa2e54172cefbeabb2+747527b0/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.1.dev0/llama/meta-llama/Llama-3.1-8B-Instruct/3e0665003fc85b7d4307.json

@@ -0,0 +1,62 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "meta-llama/Llama-3.1-8B-Instruct",
+  "_task": "text-generation",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "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": 64,
+    "capacity_factor": null,
+    "checkpoint_id": "meta-llama/Llama-3.1-8B-Instruct",
+    "checkpoint_revision": "0e9e39f249a16976918f6564b8830bc894c89659",
+    "continuous_batching": true,
+    "enable_bucketing": false,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 4,
+    "max_batch_size": 64,
+    "max_context_length": 4096,
+    "max_topk": 256,
+    "n_active_tokens": 4096,
+    "neuronxcc_version": "2.21.18209.0+043b1bf7",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.1.dev0",
+    "output_logits": false,
+    "pp_degree": 1,
+    "sequence_length": 4096,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": "trn2",
+    "torch_dtype": "bfloat16",
+    "tp_degree": 4
+  },
+  "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-2.21.18209.0+043b1bf7/MODULE_226601715d0474f1ab25+877608f3/compile_flags.json

@@ -0,0 +1 @@
+["--target=trn2", "--auto-cast=none", "--model-type=transformer", "--tensorizer-options=--enable-ccop-compute-overlap --cc-pipeline-tiling-factor=2 --vectorize-strided-dma ", "-O2", "--lnc=2", "--logfile=/tmp/nxd_model/context_encoding_model/_tp0_bk0/log-neuron-cc.txt"]

neuronxcc-2.21.18209.0+043b1bf7/MODULE_226601715d0474f1ab25+877608f3/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca5c244314061096ab5481e5ff5cf3abb8da1abcb646a24d762f2b63f58cde27
+size 2179217

neuronxcc-2.21.18209.0+043b1bf7/MODULE_226601715d0474f1ab25+877608f3/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:83338272b44fa8022a7797cf6a5c9797e48a281c98672ae8ae5c4f5d4bbcbc66
+size 3431424

neuronxcc-2.21.18209.0+043b1bf7/MODULE_fbaa2e54172cefbeabb2+747527b0/compile_flags.json

@@ -0,0 +1 @@
+["--target=trn2", "--auto-cast=none", "--model-type=transformer", "--tensorizer-options=--enable-ccop-compute-overlap --cc-pipeline-tiling-factor=2 --vectorize-strided-dma ", "-O2", "--lnc=2", "--logfile=/tmp/nxd_model/token_generation_model/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]

neuronxcc-2.21.18209.0+043b1bf7/MODULE_fbaa2e54172cefbeabb2+747527b0/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10463b56c3bc779cdf29347a9b9ad4fdc8ffc24a00653473599ae8dc676a75e0
+size 9753899

neuronxcc-2.21.18209.0+043b1bf7/MODULE_fbaa2e54172cefbeabb2+747527b0/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:397594f4a4b462a47246c9d1ada85cb6cb4a69ea8453ccdd27a7fa3046100328
+size 25785344

neuronxcc-2.21.18209.0+043b1bf7/MODULE_fbaa2e54172cefbeabb2+747527b0/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53e12d00b2defc49397e7331190ceb8dafc1080813dec3204821b9f0cddb0a26
+size 25935893