Synchronizing local compiler cache.

.gitattributes

@@ -3357,3 +3357,7 @@ neuronxcc-2.17.194.0+d312836f/MODULE_659e90ae99de96a4c99f+165e9558/wrapped_neff.
 neuronxcc-2.17.194.0+d312836f/MODULE_96d97f202e238feb2c21+bfe5714b/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.17.194.0+d312836f/MODULE_9bce7228204bd2d4577b+431f5505/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.17.194.0+d312836f/MODULE_9cd10f645f9c9838f49a+bfe5714b/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_23d27f640534fb2c4304+5be477de/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_23d27f640534fb2c4304+5be477de/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_38ed63d19a6b0761a692+84f3e719/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_9bce7228204bd2d4577b+431f5505/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.19.8089.0+8ab9f450/0_REGISTRY/0.3.0.dev3/qwen3/Qwen/Qwen3-0.6B/79fe6d5dc76e8ab41df4.json

@@ -0,0 +1,72 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "Qwen/Qwen3-0.6B",
+  "_task": "text-generation",
+  "architectures": [
+    "Qwen3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "max_position_embeddings": 40960,
+  "max_window_layers": 28,
+  "model_type": "qwen3",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "async_mode": false,
+    "attn_kernel_enabled": false,
+    "batch_size": 4,
+    "capacity_factor": null,
+    "cc_pipeline_tiling_factor": 2,
+    "checkpoint_id": "Qwen/Qwen3-0.6B",
+    "checkpoint_revision": "e6de91484c29aa9480d55605af694f39b081c455",
+    "continuous_batching": true,
+    "enable_bucketing": false,
+    "ep_degree": 1,
+    "flash_decoding_enabled": false,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "is_chunked_prefill": false,
+    "local_ranks_size": 2,
+    "logical_nc_config": 1,
+    "max_batch_size": 4,
+    "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": false,
+    "optimum_neuron_version": "0.3.0.dev3",
+    "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": 2,
+    "vocab_parallel": false
+  },
+  "num_attention_heads": 16,
+  "num_hidden_layers": 28,
+  "num_key_value_heads": 8,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000,
+  "sliding_window": null,
+  "tie_word_embeddings": true,
+  "use_cache": true,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

neuronxcc-2.19.8089.0+8ab9f450/MODULE_23d27f640534fb2c4304+5be477de/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-dge-dma --vectorize-strided-dma ", "-O2", "--internal-num-neuroncores-per-sengine=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_23d27f640534fb2c4304+5be477de/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f16d58c7f3463cf55f0d165bcf887a956b39d66b631c999c6f3aec58516b6af2
+size 694657

neuronxcc-2.19.8089.0+8ab9f450/MODULE_23d27f640534fb2c4304+5be477de/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4157630990f48fef64d2967d0fd1a8eea7f8a388b8bd51b347eebe82e3b6d826
+size 2591744

neuronxcc-2.19.8089.0+8ab9f450/MODULE_23d27f640534fb2c4304+5be477de/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d2ae9b28e5ca885ce72c31030e963f96536f5a06c1858faba85e09ea11782147
+size 2722174

neuronxcc-2.19.8089.0+8ab9f450/MODULE_38ed63d19a6b0761a692+84f3e719/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-dge-dma --vectorize-strided-dma ", "-O2", "--internal-num-neuroncores-per-sengine=1", "--logfile=/tmp/nxd_model/context_encoding_model/_tp0_bk0/log-neuron-cc.txt"]

neuronxcc-2.19.8089.0+8ab9f450/MODULE_38ed63d19a6b0761a692+84f3e719/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a7ef18d329700f84e246f207eb60b3548852b445a40ab748901d6e9775d1f3d
+size 843800

neuronxcc-2.19.8089.0+8ab9f450/MODULE_38ed63d19a6b0761a692+84f3e719/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f32c5f3bfff2efbb96a01d5dbfae90d0a67b4c326c6c349a34c8321ef43b05ae
+size 27536384

neuronxcc-2.19.8089.0+8ab9f450/MODULE_9bce7228204bd2d4577b+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_9bce7228204bd2d4577b+431f5505/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef4350dbb3041c33ed2ed3737574b565db7f7520f6fb9e34ae56e777eb6b606e
+size 132612

neuronxcc-2.19.8089.0+8ab9f450/MODULE_9bce7228204bd2d4577b+431f5505/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3eb5d1912de963a332efc0fcd15a9796f3aab8ea1a7fd56601f3f2d898ebc214
+size 492544