Synchronizing local compiler cache.

.gitattributes

@@ -11038,3 +11038,6 @@ neuronxcc-2.21.18209.0+043b1bf7/MODULE_c9f26edda1fdd24a69f2+dca7827f/model.neff
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_cdaff6e18674c5674cf6+b9128e63/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_e8c392e11918ad8bfb04+617f6939/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_e8c392e11918ad8bfb04+617f6939/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.20.9961.0+0acef03a/MODULE_4258cdff7d35fadfa642+ed72d204/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.20.9961.0+0acef03a/MODULE_ca42ff74df1fe91f342d+a9d440f5/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.20.9961.0+0acef03a/MODULE_ca42ff74df1fe91f342d+a9d440f5/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.20.9961.0+0acef03a/0_REGISTRY/0.3.1.dev6/llama/meta-llama/Llama-3.1-8B-Instruct/ca29ad0d2ab9db5dc080.json

@@ -0,0 +1,63 @@
+{
+  "_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": 1,
+    "capacity_factor": null,
+    "checkpoint_id": "meta-llama/Llama-3.1-8B-Instruct",
+    "checkpoint_revision": "0e9e39f249a16976918f6564b8830bc894c89659",
+    "continuous_batching": false,
+    "enable_bucketing": false,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 8,
+    "logical_nc_config": 1,
+    "max_batch_size": 1,
+    "max_context_length": 2048,
+    "max_topk": 256,
+    "n_active_tokens": 2048,
+    "neuronxcc_version": "2.20.9961.0+0acef03a",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.3.1.dev6",
+    "output_logits": false,
+    "pp_degree": 1,
+    "sequence_length": 2048,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": null,
+    "torch_dtype": "bfloat16",
+    "tp_degree": 8
+  },
+  "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-2.20.9961.0+0acef03a/MODULE_4258cdff7d35fadfa642+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_4258cdff7d35fadfa642+ed72d204/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7c39e4e1414870fb12198ffa8a859231eb60a109155e0b0d01303b3a2a1a3be
+size 695381

neuronxcc-2.20.9961.0+0acef03a/MODULE_4258cdff7d35fadfa642+ed72d204/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c4c1aa05f6e7d3fac39af822bbbe7ba38ca65102767d0c2b2217466c472fa6f
+size 12780544

neuronxcc-2.20.9961.0+0acef03a/MODULE_ca42ff74df1fe91f342d+a9d440f5/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/token_generation_model/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]

neuronxcc-2.20.9961.0+0acef03a/MODULE_ca42ff74df1fe91f342d+a9d440f5/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:656b8b127373025ac505a5a618cd3ae78a1513468ffa39f795dd08f5494685f1
+size 789558

neuronxcc-2.20.9961.0+0acef03a/MODULE_ca42ff74df1fe91f342d+a9d440f5/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6841884a56c7a77db59af09aaa6ef3a4921ba49560b0a712cd251fd5e316d45d
+size 1977344

neuronxcc-2.20.9961.0+0acef03a/MODULE_ca42ff74df1fe91f342d+a9d440f5/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a13bb0c147fa4d75b61cc18d7cbf0f6797bdbbc833d536670d10ad5b046b568c
+size 2115670