Synchronizing local compiler cache.

.gitattributes

@@ -7301,3 +7301,6 @@ neuronxcc-2.23.6484.0+3b612583/MODULE_a8e0dc1ba04396c0222e+ce99b21e/model.neff f
 neuronxcc-2.23.6484.0+3b612583/MODULE_a8e0dc1ba04396c0222e+ce99b21e/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.23.6484.0+3b612583/MODULE_b0287a0a8f37672e6b20+faf01bbd/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.23.6484.0+3b612583/MODULE_b0287a0a8f37672e6b20+faf01bbd/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.23.6484.0+3b612583/MODULE_2d57017f9e625202a13c+12dd9119/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.23.6484.0+3b612583/MODULE_fa3c837be0618d501d18+84337dd9/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.23.6484.0+3b612583/MODULE_fa3c837be0618d501d18+84337dd9/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.23.6484.0+3b612583/0_REGISTRY/0.4.6.dev3/cf6b9a360dcf294104671106bae2adbd9fd291823bb60a351883163684073231/d6cc94cca9a466f07ad8.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": "2.23.6484.0+3b612583",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.6.dev3",
+    "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-2.23.6484.0+3b612583/MODULE_2d57017f9e625202a13c+12dd9119/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", "--verbose=35", "--logfile=/tmp/nxd_model/chunked_prefill/_tp0_bk0/log-neuron-cc.txt"]

neuronxcc-2.23.6484.0+3b612583/MODULE_2d57017f9e625202a13c+12dd9119/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:552b0fd83561286728742079aa65c0b30d4e154aa6dfac4b849c951302efb74e
+size 1818393

neuronxcc-2.23.6484.0+3b612583/MODULE_2d57017f9e625202a13c+12dd9119/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ae56679439cdd6c19e71daeba509b20082935777670209637532ea7a01aa750
+size 12053504

neuronxcc-2.23.6484.0+3b612583/MODULE_fa3c837be0618d501d18+84337dd9/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", "--verbose=35", "--logfile=/tmp/nxd_model/token_generation/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]

neuronxcc-2.23.6484.0+3b612583/MODULE_fa3c837be0618d501d18+84337dd9/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da5f1f83e0569f52234f3e21088d59679367d6e782a4e9534ae1d107b6cf7705
+size 1418893

neuronxcc-2.23.6484.0+3b612583/MODULE_fa3c837be0618d501d18+84337dd9/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4d528b66afd11339800e233f57f9dc20ec666bd3822cc4ecb2e4840998bc005
+size 2735104

neuronxcc-2.23.6484.0+3b612583/MODULE_fa3c837be0618d501d18+84337dd9/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8673d4b9496352dec711887e29d7bd179cc45bf40302fde67e28a877daf016da
+size 2809191