Synchronizing local compiler cache.

.gitattributes

@@ -15348,3 +15348,6 @@ neuronxcc-2.21.33363.0+82129205/MODULE_ebdaf819ea1c1978f2f3+a32116a7/model.neff
 neuronxcc-2.21.33363.0+82129205/MODULE_41a4241c31add0926b42+a02c3a36/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_41a4241c31add0926b42+a02c3a36/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_513ab40850afac85787a+24129607/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_28637e1e5d207b352ec3+a32116a7/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_f9bebcec047382c8e547+ac10809c/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_f9bebcec047382c8e547+ac10809c/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.4.dev2/llama/meta-llama/Llama-3.1-8B/9a073ea660d7a743b8d0.json

@@ -0,0 +1,62 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "meta-llama/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": 64,
+    "capacity_factor": null,
+    "checkpoint_id": "meta-llama/Llama-3.1-8B",
+    "checkpoint_revision": "d04e592bb4f6aa9cfee91e2e20afa771667e1d4b",
+    "continuous_batching": true,
+    "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": "2.21.33363.0+82129205",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.4.dev2",
+    "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-2.21.33363.0+82129205/MODULE_28637e1e5d207b352ec3+a32116a7/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/_tp0_bk0/log-neuron-cc.txt"]

neuronxcc-2.21.33363.0+82129205/MODULE_28637e1e5d207b352ec3+a32116a7/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc4108ecdf261bac439769baca25275b7b303c337d1e861b4304da93da0e48fe
+size 2204489

neuronxcc-2.21.33363.0+82129205/MODULE_28637e1e5d207b352ec3+a32116a7/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f310e9e6b69de8c2fa46628b15e79716ec7aa803d89e0a59af7e580f49dd293a
+size 3431424

neuronxcc-2.21.33363.0+82129205/MODULE_f9bebcec047382c8e547+ac10809c/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/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]

neuronxcc-2.21.33363.0+82129205/MODULE_f9bebcec047382c8e547+ac10809c/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6620f5cb3d32d6be6140ea2fff3a5b96541a41769b3ebdc461787b673bbfe8de
+size 9772123

neuronxcc-2.21.33363.0+82129205/MODULE_f9bebcec047382c8e547+ac10809c/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4b26104316a71f65912e7a76231a2bbb9df507e8055901476670d58ed377e3df
+size 25816064

neuronxcc-2.21.33363.0+82129205/MODULE_f9bebcec047382c8e547+ac10809c/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc1420b792dbca8e7a515232e5b9dbd87c8deb9626c833d53d7f2e90f7cd41b8
+size 25966470