Synchronizing local compiler cache.

.gitattributes

@@ -13664,3 +13664,6 @@ neuronxcc-2.21.33363.0+82129205/MODULE_cc50f4afe3c0144e7f0f+24129607/model.neff
 neuronxcc-2.21.33363.0+82129205/MODULE_5c79a551d80767988316+a02c3a36/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_5c79a551d80767988316+a02c3a36/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_6c50a0e193cbaee10f4e+24129607/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_d6903651a714fe104a12+24129607/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_fd811dfabff9a728f487+a02c3a36/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_fd811dfabff9a728f487+a02c3a36/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.3.dev2/llama/meta-llama/Llama-2-7b-hf/2374859c01d4d11abce0.json

@@ -0,0 +1,56 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "meta-llama/Llama-2-7b-hf",
+  "_task": "text-generation",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "dtype": "float16",
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 4096,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 1,
+    "capacity_factor": null,
+    "checkpoint_id": "meta-llama/Llama-2-7b-hf",
+    "checkpoint_revision": "01c7f73d771dfac7d292323805ebc428287df4f9",
+    "continuous_batching": false,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 1,
+    "max_context_length": 2048,
+    "max_topk": 256,
+    "n_active_tokens": 2048,
+    "neuronxcc_version": "2.21.33363.0+82129205",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.3.dev2",
+    "output_logits": false,
+    "pp_degree": 1,
+    "sequence_length": 2048,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": "trn1",
+    "torch_dtype": "float16",
+    "tp_degree": 2
+  },
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 32,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "use_cache": true,
+  "vocab_size": 32000
+}

neuronxcc-2.21.33363.0+82129205/MODULE_d6903651a714fe104a12+24129607/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/_tp0_bk0/log-neuron-cc.txt"]

neuronxcc-2.21.33363.0+82129205/MODULE_d6903651a714fe104a12+24129607/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a3b9cc34850bcc619c0e789a108b17ea5c6d84b3faa1171342dbfcf2515d430b
+size 653427

neuronxcc-2.21.33363.0+82129205/MODULE_d6903651a714fe104a12+24129607/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:06318806f174740939a407b1214d8662fae49e21c59b38f02ef5bc91e322a7fd
+size 2857984

neuronxcc-2.21.33363.0+82129205/MODULE_fd811dfabff9a728f487+a02c3a36/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/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]

neuronxcc-2.21.33363.0+82129205/MODULE_fd811dfabff9a728f487+a02c3a36/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e733fd2f3adfdda470b51b1296911af7b4243880c076d8944f842e324f32583a
+size 694440

neuronxcc-2.21.33363.0+82129205/MODULE_fd811dfabff9a728f487+a02c3a36/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:29ebb7f3c287c60a9fd07c048c6850f9143f44efe24ff05f9c2fdfd56da5e445
+size 10937344

neuronxcc-2.21.33363.0+82129205/MODULE_fd811dfabff9a728f487+a02c3a36/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:270c80c812161f7472fed833e123b1ff0853de141ecc4f2e1c3df2cc2d027c89
+size 11085182