Synchronizing local compiler cache.

.gitattributes

@@ -18623,3 +18623,6 @@ neuronxcc-2.26.6360.0+6f180f47/MODULE_ea5da79b2b0ee89bf62e+4a27301b/wrapped_neff
 neuronxcc-2.26.6360.0+6f180f47/MODULE_493f9a6002edf032518c+84337dd9/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.26.6360.0+6f180f47/MODULE_493f9a6002edf032518c+84337dd9/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.26.6360.0+6f180f47/MODULE_faae9542f2a0e7129ab1+519f203d/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.26.6360.0+6f180f47/MODULE_959c1462780e286d152a+84337dd9/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.26.6360.0+6f180f47/MODULE_959c1462780e286d152a+84337dd9/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.26.6360.0+6f180f47/MODULE_c3cd06b8f339dad5453a+519f203d/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.26.6360.0+6f180f47/0_REGISTRY/0.4.7.dev0/040f4a4a5adbaf01ad45deee70f0d520147ff8515b9c5011c11300eb834c3375/d3feb02c3903aef652be.json

@@ -0,0 +1,64 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "llamafactory/tiny-random-qwen3",
+  "_task": "text-generation",
+  "architectures": [
+    "Qwen3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "dtype": "bfloat16",
+  "head_dim": 4,
+  "hidden_act": "silu",
+  "hidden_size": 16,
+  "initializer_range": 0.02,
+  "intermediate_size": 64,
+  "layer_types": [
+    "full_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 32768,
+  "max_window_layers": 36,
+  "model_type": "qwen3",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 2,
+    "capacity_factor": null,
+    "checkpoint_id": "llamafactory/tiny-random-qwen3",
+    "checkpoint_revision": "81d6f5f5e05ed53ea8a1d19431266a486e46bbd8",
+    "continuous_batching": true,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 2,
+    "max_context_length": 1024,
+    "max_topk": 256,
+    "n_active_tokens": 1024,
+    "neuronxcc_version": "2.26.6360.0+6f180f47",
+    "on_device_sampling": false,
+    "optimum_neuron_version": "0.4.7.dev0",
+    "output_logits": false,
+    "pp_degree": 1,
+    "prefill_chunk_size": 0,
+    "sequence_length": 1024,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": "trn1",
+    "torch_dtype": "bfloat16",
+    "tp_degree": 2
+  },
+  "num_attention_heads": 4,
+  "num_hidden_layers": 2,
+  "num_key_value_heads": 4,
+  "rms_norm_eps": 1e-06,
+  "rope_parameters": {
+    "rope_theta": 5000000,
+    "rope_type": "default"
+  },
+  "sliding_window": null,
+  "tie_word_embeddings": true,
+  "use_cache": true,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

neuronxcc-2.26.6360.0+6f180f47/0_REGISTRY/0.4.7.dev0/qwen3/llamafactory/tiny-random-qwen3/d3feb02c3903aef652be.json

@@ -0,0 +1,64 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "llamafactory/tiny-random-qwen3",
+  "_task": "text-generation",
+  "architectures": [
+    "Qwen3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "dtype": "bfloat16",
+  "head_dim": 4,
+  "hidden_act": "silu",
+  "hidden_size": 16,
+  "initializer_range": 0.02,
+  "intermediate_size": 64,
+  "layer_types": [
+    "full_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 32768,
+  "max_window_layers": 36,
+  "model_type": "qwen3",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 2,
+    "capacity_factor": null,
+    "checkpoint_id": "llamafactory/tiny-random-qwen3",
+    "checkpoint_revision": "81d6f5f5e05ed53ea8a1d19431266a486e46bbd8",
+    "continuous_batching": true,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 2,
+    "max_context_length": 1024,
+    "max_topk": 256,
+    "n_active_tokens": 1024,
+    "neuronxcc_version": "2.26.6360.0+6f180f47",
+    "on_device_sampling": false,
+    "optimum_neuron_version": "0.4.7.dev0",
+    "output_logits": false,
+    "pp_degree": 1,
+    "prefill_chunk_size": 0,
+    "sequence_length": 1024,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": "trn1",
+    "torch_dtype": "bfloat16",
+    "tp_degree": 2
+  },
+  "num_attention_heads": 4,
+  "num_hidden_layers": 2,
+  "num_key_value_heads": 4,
+  "rms_norm_eps": 1e-06,
+  "rope_parameters": {
+    "rope_theta": 5000000,
+    "rope_type": "default"
+  },
+  "sliding_window": null,
+  "tie_word_embeddings": true,
+  "use_cache": true,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

neuronxcc-2.26.6360.0+6f180f47/MODULE_959c1462780e286d152a+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-2.26.6360.0+6f180f47/MODULE_959c1462780e286d152a+84337dd9/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f0cb30ba9e46be44cba660c4312f51c9ced939e5906823e6e7ec2294342bcd60
+size 51953

neuronxcc-2.26.6360.0+6f180f47/MODULE_959c1462780e286d152a+84337dd9/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5e89849342964b90d04eec737c401deae0a10406d7fbcb8fedda533b07a0716
+size 277504

neuronxcc-2.26.6360.0+6f180f47/MODULE_959c1462780e286d152a+84337dd9/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:403e334010cbb27a3d8c8272e23a4f92616cd86f494962ebdfdf5df03ca0ff51
+size 285649

neuronxcc-2.26.6360.0+6f180f47/MODULE_c3cd06b8f339dad5453a+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-2.26.6360.0+6f180f47/MODULE_c3cd06b8f339dad5453a+519f203d/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c8c267d5fbe96157bed49d29a863a35e4cae8a3f1c9ef92a492edaf58aca0bb2
+size 51303

neuronxcc-2.26.6360.0+6f180f47/MODULE_c3cd06b8f339dad5453a+519f203d/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b122ae76f46b4f36b0a2b210b60f1dcf54d4e165ba91e1ef7ac30343ff241938
+size 297984