Synchronizing local compiler cache.

.gitattributes

@@ -9548,7 +9548,7 @@ neuronxcc-2.19.8089.0+8ab9f450/MODULE_18bf9b0d58fe3b48bade+a9d440f5/model.neff f
 neuronxcc-2.19.8089.0+8ab9f450/MODULE_18bf9b0d58fe3b48bade+a9d440f5/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.19.8089.0+8ab9f450/MODULE_a371a0fbf9933f3dfaf4+ed72d204/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.19.8089.0+8ab9f450/MODULE_b3e2c91b97f0eb0f9d3b+431f5505/model.neff filter=lfs diff=lfs merge=lfs -text
-neuronxcc-2.19.8089.0+8ab9f450/MODULE_7b42b317dc46bf6976d7+a9d440f5/model.neff filter=lfs diff=lfs merge=lfs -text
-neuronxcc-2.19.8089.0+8ab9f450/MODULE_7b42b317dc46bf6976d7+a9d440f5/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
-neuronxcc-2.19.8089.0+8ab9f450/MODULE_902c21c3373aee223123+431f5505/model.neff filter=lfs diff=lfs merge=lfs -text
-neuronxcc-2.19.8089.0+8ab9f450/MODULE_f2b825d881db5c86afd5+ed72d204/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_1da9172c2c2642d60cd0+a9d440f5/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_1da9172c2c2642d60cd0+a9d440f5/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_67733bc879ab1bd682c4+431f5505/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_b9c9280f6734f291e466+ed72d204/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.19.8089.0+8ab9f450/0_REGISTRY/0.3.0.dev5/llama/meta-llama/Llama-2-7b-hf/91d3b8109c503d71dd94.json

@@ -0,0 +1,71 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "meta-llama/Llama-2-7b-hf",
+  "_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": 11008,
+  "max_position_embeddings": 4096,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "async_mode": false,
+    "attn_kernel_enabled": false,
+    "batch_size": 1,
+    "capacity_factor": null,
+    "cc_pipeline_tiling_factor": 2,
+    "checkpoint_id": "meta-llama/Llama-2-7b-hf",
+    "checkpoint_revision": "01c7f73d771dfac7d292323805ebc428287df4f9",
+    "continuous_batching": false,
+    "enable_bucketing": false,
+    "ep_degree": 1,
+    "flash_decoding_enabled": false,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "is_chunked_prefill": false,
+    "local_ranks_size": 2,
+    "logical_nc_config": 1,
+    "max_batch_size": 1,
+    "max_context_length": 2048,
+    "max_topk": 256,
+    "mlp_kernel_enabled": false,
+    "mlp_kernel_fuse_residual_add": false,
+    "n_active_tokens": 2048,
+    "neuronxcc_version": "2.19.8089.0+8ab9f450",
+    "num_cores_per_group": 1,
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.3.0.dev5",
+    "output_logits": false,
+    "padding_side": "right",
+    "pp_degree": 1,
+    "qk_layernorm": false,
+    "qkv_kernel_enabled": false,
+    "rpl_reduce_dtype": "bfloat16",
+    "sequence_length": 2048,
+    "sequence_parallel_enabled": false,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": null,
+    "torch_dtype": "bfloat16",
+    "tp_degree": 2,
+    "vocab_parallel": false
+  },
+  "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-2.19.8089.0+8ab9f450/MODULE_1da9172c2c2642d60cd0+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-2.19.8089.0+8ab9f450/MODULE_1da9172c2c2642d60cd0+a9d440f5/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d81489d282fb8caeb02c398bb0b266c309e803689694a0230eb65af06951e77
+size 692063

neuronxcc-2.19.8089.0+8ab9f450/MODULE_1da9172c2c2642d60cd0+a9d440f5/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:00bc644883c006573b38e8bb238bf66f307de38c851cca6c8534c7f30c1835ed
+size 8049664

neuronxcc-2.19.8089.0+8ab9f450/MODULE_1da9172c2c2642d60cd0+a9d440f5/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b2c5170b1ae1701f1dc8f476e0a1754925f58a99139a34439d073ef639291d60
+size 8188006

neuronxcc-2.19.8089.0+8ab9f450/MODULE_67733bc879ab1bd682c4+431f5505/compile_flags.json

@@ -0,0 +1 @@
+"--model-type=transformer -O1 --lnc=1 --internal-hlo2tensorizer-options=--experimental-unsafe-fp8e4m3fn-as-fp8e4m3 --logfile=/tmp/nxd_model/layout_opt/log-neuron-cc.txt"

neuronxcc-2.19.8089.0+8ab9f450/MODULE_67733bc879ab1bd682c4+431f5505/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53fcfe474a391b2722e80464060f90172f76e7fe4e3ba2efd13d76cfddb95a64
+size 136732

neuronxcc-2.19.8089.0+8ab9f450/MODULE_67733bc879ab1bd682c4+431f5505/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1881d9888d549aa921f014906d145a1ef7c6fc73d0b73e0cfc9ba3abaaef8b1b
+size 7015424

neuronxcc-2.19.8089.0+8ab9f450/MODULE_b9c9280f6734f291e466+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-2.19.8089.0+8ab9f450/MODULE_b9c9280f6734f291e466+ed72d204/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d4eb818726e3992be8ccedb929dd4b0a8d369024fe11fb1a71ec50b09ae3cde
+size 625305

neuronxcc-2.19.8089.0+8ab9f450/MODULE_b9c9280f6734f291e466+ed72d204/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:352c77c2ac6d0f578943d610597f30d4a2382ab9e15d0c09c72af7a6426b9dfd
+size 3318784