Synchronizing local compiler cache.

.gitattributes

@@ -5901,3 +5901,6 @@ neuronxcc-2.21.33363.0+82129205/MODULE_9fe6e286c1c48b3cf2a9+a02c3a36/wrapped_nef
 neuronxcc-2.21.33363.0+82129205/MODULE_bd1e72b121bb7318c283+a02c3a36/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_bd1e72b121bb7318c283+a02c3a36/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_f1dce126097f38c8bbe1+24129607/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.22.12471.0+b4a00d10/MODULE_5274831e83842758b831+519f203d/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.22.12471.0+b4a00d10/MODULE_a968bbe6dc74ca5087a2+84337dd9/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.22.12471.0+b4a00d10/MODULE_a968bbe6dc74ca5087a2+84337dd9/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.22.12471.0+b4a00d10/0_REGISTRY/0.4.4.dev2/llama/unsloth/Llama-3.2-1B-Instruct/7cba097179296102aa62.json

@@ -0,0 +1,63 @@
+{
+  "_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": 4,
+    "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": 4,
+    "max_context_length": 4096,
+    "max_topk": 256,
+    "n_active_tokens": 4096,
+    "neuronxcc_version": "2.22.12471.0+b4a00d10",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.4.dev2",
+    "output_logits": false,
+    "pp_degree": 1,
+    "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-2.22.12471.0+b4a00d10/MODULE_5274831e83842758b831+519f203d/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/context_encoding/_tp0_bk0/log-neuron-cc.txt"]

neuronxcc-2.22.12471.0+b4a00d10/MODULE_5274831e83842758b831+519f203d/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bce5a5ee8747ce6e141cdff4b83348032f3268e73c644366ad0b5a5dd7b417a2
+size 1511489

neuronxcc-2.22.12471.0+b4a00d10/MODULE_5274831e83842758b831+519f203d/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5df396b699dc54f4e90f1bf7132b6f2bb5dbcc31630cb56c922ffa7badde4416
+size 4219904

neuronxcc-2.22.12471.0+b4a00d10/MODULE_a968bbe6dc74ca5087a2+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-2.22.12471.0+b4a00d10/MODULE_a968bbe6dc74ca5087a2+84337dd9/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:39648c1c623205b5083818c2674038e906b70d09df0442cc757e1f6ac36e7607
+size 1722032

neuronxcc-2.22.12471.0+b4a00d10/MODULE_a968bbe6dc74ca5087a2+84337dd9/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c87e8fa425f3a6eee0b67b1dc4657a91a4598686e5b0ad9a1b1b8813f61dc24
+size 3257344

neuronxcc-2.22.12471.0+b4a00d10/MODULE_a968bbe6dc74ca5087a2+84337dd9/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e41012fda7d4998c101efde3a213f17492ad42975b34fb1dbbbac8b7038f6ffd
+size 3331655