Synchronizing local compiler cache.

.gitattributes

@@ -11350,3 +11350,6 @@ neuronxcc-2.21.18209.0+043b1bf7/MODULE_ab069b6ad71a98a912f0+a9d440f5/wrapped_nef
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_55c74c8bdcca24f67606+ed72d204/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_e93122cb41a3e760e36f+a9d440f5/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_e93122cb41a3e760e36f+a9d440f5/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_6da5b445e72c8475b566+ed72d204/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_cda5f1e1c04b9498da62+a9d440f5/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_cda5f1e1c04b9498da62+a9d440f5/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.3.1.dev7/llama/princeton-nlp/Sheared-LLaMA-1.3B/07739d60d9e8a2749b85.json

@@ -0,0 +1,57 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "princeton-nlp/Sheared-LLaMA-1.3B",
+  "_task": "text-generation",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 5504,
+  "max_position_embeddings": 4096,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 1,
+    "capacity_factor": null,
+    "checkpoint_id": "princeton-nlp/Sheared-LLaMA-1.3B",
+    "checkpoint_revision": "a4b76938edbf571ea7d7d9904861cbdca08809b4",
+    "continuous_batching": false,
+    "enable_bucketing": false,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "logical_nc_config": 1,
+    "max_batch_size": 1,
+    "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": "0.3.1.dev7",
+    "output_logits": false,
+    "pp_degree": 1,
+    "sequence_length": 4096,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": null,
+    "torch_dtype": "float16",
+    "tp_degree": 2
+  },
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "num_key_value_heads": 16,
+  "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-2.21.18209.0+043b1bf7/MODULE_6da5b445e72c8475b566+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-2.21.18209.0+043b1bf7/MODULE_6da5b445e72c8475b566+ed72d204/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6613a12de7f842a3d476f38f7449049c8d0fd3ded59d8510caf763e29d0f1e3f
+size 654134

neuronxcc-2.21.18209.0+043b1bf7/MODULE_6da5b445e72c8475b566+ed72d204/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d7da2c60aea3248948fc965906e84bd932e686675fb99ee1239d1a87600f3e4
+size 42814464

neuronxcc-2.21.18209.0+043b1bf7/MODULE_cda5f1e1c04b9498da62+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-2.21.18209.0+043b1bf7/MODULE_cda5f1e1c04b9498da62+a9d440f5/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e80f1f80dd8ac75158c1d26290b5b47ddf40db086bfe6dc1d96bcde265cd21c
+size 531329

neuronxcc-2.21.18209.0+043b1bf7/MODULE_cda5f1e1c04b9498da62+a9d440f5/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c6be5e030a10fbf0ed13b946d87806b1dbfeacc2b71dfb064c88b03bd7b3735
+size 2817024

neuronxcc-2.21.18209.0+043b1bf7/MODULE_cda5f1e1c04b9498da62+a9d440f5/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1bfb799044227eb2662d8beca05aa45d8a6f6982083be4a1a5520a093aaeb439
+size 2927540

neuronxcc-2.21.18209.0+043b1bf7/MODULE_e5881ad1904fef704fb7+a9d440f5/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5232114001bd4441991560bb14c60dc448174296ba57a9bf540da71f1c229c16
+oid sha256:863d71a8785735462742c9cfad7e18bd559721d9dcc9b03eb02ec51d488f666e
 size 6738944

neuronxcc-2.21.18209.0+043b1bf7/MODULE_e5881ad1904fef704fb7+a9d440f5/wrapped_neff.hlo

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e75fef7f30ab88b32f259fa4b5db1332cd4eb5883d94dc3de9580d1c4e7ef409
+oid sha256:e7ef6c50222c9efd5ba6cd79fc8a2a3e85cdc9b816e9b7e65de7d3e9791ae51f
 size 6886551