Synchronizing local compiler cache.

.gitattributes

@@ -9440,3 +9440,7 @@ neuronxcc-2.19.8089.0+8ab9f450/MODULE_3baa91d1c83900f58dfa+ed72d204/model.neff f
 neuronxcc-2.19.8089.0+8ab9f450/MODULE_4ea1c50770caad6065ae+a9d440f5/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.19.8089.0+8ab9f450/MODULE_4ea1c50770caad6065ae+a9d440f5/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.19.8089.0+8ab9f450/MODULE_55997df3963c10393bae+431f5505/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_53f685ebc1a83f7e8774+ed72d204/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_58c742d10c979483b114+431f5505/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_90c4af9af37cec5a5f50+a9d440f5/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_90c4af9af37cec5a5f50+a9d440f5/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.19.8089.0+8ab9f450/0_REGISTRY/0.3.0.dev5/qwen2/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B/9f6889b9644215c4765d.json

@@ -0,0 +1,70 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
+  "_task": "text-generation",
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "hidden_act": "silu",
+  "hidden_size": 5120,
+  "initializer_range": 0.02,
+  "intermediate_size": 27648,
+  "max_position_embeddings": 131072,
+  "max_window_layers": 64,
+  "model_type": "qwen2",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "async_mode": false,
+    "attn_kernel_enabled": false,
+    "batch_size": 1,
+    "capacity_factor": null,
+    "cc_pipeline_tiling_factor": 2,
+    "checkpoint_id": "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
+    "checkpoint_revision": "711ad2ea6aa40cfca18895e8aca02ab92df1a746",
+    "continuous_batching": false,
+    "enable_bucketing": false,
+    "ep_degree": 1,
+    "flash_decoding_enabled": false,
+    "fused_qkv": false,
+    "glu_mlp": true,
+    "is_chunked_prefill": false,
+    "local_ranks_size": 8,
+    "logical_nc_config": 1,
+    "max_batch_size": 1,
+    "max_context_length": 4096,
+    "max_topk": 256,
+    "mlp_kernel_enabled": false,
+    "mlp_kernel_fuse_residual_add": false,
+    "n_active_tokens": 4096,
+    "neuronxcc_version": "2.19.8089.0+8ab9f450",
+    "num_cores_per_group": 1,
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.3.0.dev5",
+    "output_logits": false,
+    "padding_side": "right",
+    "pp_degree": 1,
+    "qk_layernorm": false,
+    "qkv_kernel_enabled": false,
+    "rpl_reduce_dtype": "bfloat16",
+    "sequence_length": 4096,
+    "sequence_parallel_enabled": false,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": null,
+    "torch_dtype": "bfloat16",
+    "tp_degree": 8,
+    "vocab_parallel": false
+  },
+  "num_attention_heads": 40,
+  "num_hidden_layers": 64,
+  "num_key_value_heads": 8,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 1000000.0,
+  "sliding_window": 131072,
+  "tie_word_embeddings": false,
+  "use_cache": true,
+  "use_sliding_window": false,
+  "vocab_size": 152064
+}

neuronxcc-2.19.8089.0+8ab9f450/MODULE_53f685ebc1a83f7e8774+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-2.19.8089.0+8ab9f450/MODULE_53f685ebc1a83f7e8774+ed72d204/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d91825f01bb9049dc72b3d29a572ad399361bf19232da4b17e7fbddb996a77cf
+size 1837579

neuronxcc-2.19.8089.0+8ab9f450/MODULE_53f685ebc1a83f7e8774+ed72d204/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f80053b2650173fcc21329edddbf102f4baf950e31a20dc2860cc219a614776d
+size 4260864

neuronxcc-2.19.8089.0+8ab9f450/MODULE_58c742d10c979483b114+431f5505/compile_flags.json

@@ -0,0 +1 @@
+"--model-type=transformer -O1 --lnc=1 --internal-hlo2tensorizer-options=--experimental-unsafe-fp8e4m3fn-as-fp8e4m3 --logfile=/tmp/nxd_model/layout_opt/log-neuron-cc.txt"

neuronxcc-2.19.8089.0+8ab9f450/MODULE_58c742d10c979483b114+431f5505/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f14432af9f2c69e8216e92146a3b47a5167f540f2f3c06db568c95d2c1380b1
+size 411300

neuronxcc-2.19.8089.0+8ab9f450/MODULE_58c742d10c979483b114+431f5505/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e452a97b55524499e1d681c184798dbc342a7e972f56c37aa8b860d604d1cba
+size 7250944

neuronxcc-2.19.8089.0+8ab9f450/MODULE_90c4af9af37cec5a5f50+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-2.19.8089.0+8ab9f450/MODULE_90c4af9af37cec5a5f50+a9d440f5/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c281cd3426792a2acf292f5429e0f9c44f0b96ee6c0940b8f098f9966f00108d
+size 1623157

neuronxcc-2.19.8089.0+8ab9f450/MODULE_90c4af9af37cec5a5f50+a9d440f5/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25366fc935ca9be9868c20c91d4a73a43863b3f2183ac7c2b51ca3656da4ae9a
+size 6605824

neuronxcc-2.19.8089.0+8ab9f450/MODULE_90c4af9af37cec5a5f50+a9d440f5/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bfb0aca7be5dafac2dcb6ed40a04255d09d976030acc3f3a865b6b2c06f4cb56
+size 7001046