Synchronizing local compiler cache.

.gitattributes

@@ -7109,3 +7109,5 @@ neuronxcc-2.21.33363.0+82129205/MODULE_ef93001ee65461791427+24129607/model.neff
 neuronxcc-2.21.33363.0+82129205/MODULE_f33d66622d796b2877aa+283df001/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_fa29fc69337939622d95+24129607/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_fb598b8eba6e5a92e75f+24129607/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_19d536f91de316380360+a02c3a36/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_19d536f91de316380360+a02c3a36/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev2/cf6b9a360dcf294104671106bae2adbd9fd291823bb60a351883163684073231/66850ee023afb10c2caa.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": 2,
+    "capacity_factor": null,
+    "checkpoint_id": "unsloth/Llama-3.2-1B-Instruct",
+    "checkpoint_revision": "5a8abab4a5d6f164389b1079fb721cfab8d7126c",
+    "continuous_batching": true,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 2,
+    "max_context_length": 4096,
+    "max_topk": 256,
+    "n_active_tokens": 4096,
+    "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": 0,
+    "sequence_length": 4096,
+    "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_19d536f91de316380360+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_19d536f91de316380360+a02c3a36/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c845502e4ef9fa82d65d70b52c312f34aca3ee077e79470b1afa3ca8a2f7c49
+size 443823

neuronxcc-2.21.33363.0+82129205/MODULE_19d536f91de316380360+a02c3a36/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5e07b947bf0abfc6bc9707aa7457e63aa1922ba7060bc1ba08f865e2464d463
+size 2581504

neuronxcc-2.21.33363.0+82129205/MODULE_19d536f91de316380360+a02c3a36/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f7b9a37b89e692cc36e0ea4e413da0687e4c0af0b29463c8dae4d697927b700
+size 2655591

neuronxcc-2.21.33363.0+82129205/MODULE_ad0a77ae55980efcab6b+24129607/model.hlo_module.pb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c05ad527dd98530ff1934cc0978bb11d88331152209644114965f2b45a6643fe
+oid sha256:c65609a71cbe4ecb16e918e6f3330aa44cbd761df88bbe65933076e7f029f7c0
 size 509366

neuronxcc-2.21.33363.0+82129205/MODULE_ad0a77ae55980efcab6b+24129607/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:19e0fc56822f8939bf4339e34cfc3138ec4a04a0fea224b58242ee952097ec84
+oid sha256:e7ce56562e15177aeb16cdcdda139cbd4630ec237a8e4af56281b946838311ba
 size 41616384