Synchronizing local compiler cache.

.gitattributes

@@ -7115,3 +7115,4 @@ neuronxcc-2.21.33363.0+82129205/MODULE_bc1c29d0ae34bc7587d9+6170d8e1/model.neff
 neuronxcc-2.21.33363.0+82129205/MODULE_691f0ef9fca273527dbe+24129607/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_a3c6a617f4311e09ad49+a02c3a36/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_a3c6a617f4311e09ad49+a02c3a36/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_ab293424ba3cf7b684c0+6170d8e1/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev2/cf6b9a360dcf294104671106bae2adbd9fd291823bb60a351883163684073231/0de3c6ac01567e192255.json

@@ -0,0 +1,64 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "unsloth/Llama-3.2-1B-Instruct",
+  "_task": "text-generation",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "dtype": "bfloat16",
+  "head_dim": 64,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "max_position_embeddings": 131072,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 1,
+    "capacity_factor": null,
+    "checkpoint_id": "unsloth/Llama-3.2-1B-Instruct",
+    "checkpoint_revision": "5a8abab4a5d6f164389b1079fb721cfab8d7126c",
+    "continuous_batching": false,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 1,
+    "max_context_length": 8192,
+    "max_topk": 256,
+    "n_active_tokens": 8192,
+    "neuronxcc_version": "2.21.33363.0+82129205",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.6.dev2",
+    "output_logits": false,
+    "pp_degree": 1,
+    "prefill_chunk_size": 1024,
+    "sequence_length": 8192,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": "trn1",
+    "torch_dtype": "bfloat16",
+    "tp_degree": 2
+  },
+  "num_attention_heads": 32,
+  "num_hidden_layers": 16,
+  "num_key_value_heads": 8,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": {
+    "factor": 32.0,
+    "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": true,
+  "unsloth_fixed": true,
+  "use_cache": true,
+  "vocab_size": 128256
+}

neuronxcc-2.21.33363.0+82129205/MODULE_95b9072f246645f24461+a02c3a36/model.hlo_module.pb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:12bcc80b2ad2cb1e12abbca61cd3f324f6f4f564cf78d121d0f5fcd01305c683
+oid sha256:b364b15f88cdba27d3ac4973f7dcfb2cbc49da9abd971bc1f0c52315364d55dc
 size 451319

neuronxcc-2.21.33363.0+82129205/MODULE_95b9072f246645f24461+a02c3a36/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bdf04a70eb8bda2f7f4e49bedb2c346136c7e9b0dec32fa7421ea72474563fbe
+oid sha256:4e6c30b74af481d02526f6f7d1de0fae7d17229a1fa9933e4f44b2d227190621
 size 2509824

neuronxcc-2.21.33363.0+82129205/MODULE_95b9072f246645f24461+a02c3a36/wrapped_neff.hlo

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c84f3a91584fb71a18f3ba16dd9c1cde9917ff28ab72e432ab03a61ad722f8dc
+oid sha256:4241cf6c8251d957dffdda2764ea64de989cfd720460b4dbced565d4676b2033
 size 2583911

neuronxcc-2.21.33363.0+82129205/MODULE_ab293424ba3cf7b684c0+6170d8e1/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/chunked_prefill/_tp0_bk0/log-neuron-cc.txt"]

neuronxcc-2.21.33363.0+82129205/MODULE_ab293424ba3cf7b684c0+6170d8e1/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ccc4d786d555106ebf055b55dfd76562bcdfd7cad3dcadb757119bc5ddd14bc7
+size 559974

neuronxcc-2.21.33363.0+82129205/MODULE_ab293424ba3cf7b684c0+6170d8e1/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48e52fd14c2a297a3f9ada58dca39809587b2b46b4f36740b34b320a016ac491
+size 44893184