Synchronizing local compiler cache.

.gitattributes

@@ -2780,3 +2780,5 @@ neuronxcc-2.18.121.0+9e31e41a/MODULE_e9bcfc17d832317203bd+613edded/model.neff fi
 neuronxcc-2.18.121.0+9e31e41a/MODULE_f91cd98a64f373af274d+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.18.121.0+9e31e41a/MODULE_fce469267b2ad1b5d80e+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.18.121.0+9e31e41a/MODULE_fda7fb53c475ba393ed7+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.18.121.0+9e31e41a/MODULE_2e229618015e416964c5+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.18.121.0+9e31e41a/MODULE_4b31f40da9553c6566ed+613edded/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.18.121.0+9e31e41a/0_REGISTRY/0.2.0.dev7/qwen2/Qwen/Qwen2.5-0.5B/71f4147eae9e0705be9c.json

@@ -0,0 +1,49 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "Qwen/Qwen2.5-0.5B",
+  "_task": "text-generation",
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "hidden_act": "silu",
+  "hidden_size": 896,
+  "initializer_range": 0.02,
+  "intermediate_size": 4864,
+  "max_position_embeddings": 32768,
+  "max_window_layers": 24,
+  "model_type": "qwen2",
+  "neuron": {
+    "_serialized_key": "HloNeuronConfig",
+    "all_reduce_dtype": null,
+    "allow_flash_attention": true,
+    "attention_layout": "HSB",
+    "attn_output_transposed": false,
+    "auto_cast_type": "fp16",
+    "batch_size": 4,
+    "checkpoint_id": "Qwen/Qwen2.5-0.5B",
+    "checkpoint_revision": "060db6499f32faf8b98477b0a26969ef7d8b9987",
+    "collectives_layout": "HSB",
+    "continuous_batching": true,
+    "fuse_qkv": false,
+    "group_query_attention": "shard-over-heads",
+    "log_softmax_scores": false,
+    "neuronxcc_version": "2.18.121.0+9e31e41a",
+    "optimum_neuron_version": "0.2.0.dev7",
+    "output_all_logits": false,
+    "sequence_length": 4096,
+    "tp_degree": 2
+  },
+  "num_attention_heads": 14,
+  "num_hidden_layers": 24,
+  "num_key_value_heads": 2,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000.0,
+  "sliding_window": 32768,
+  "tie_word_embeddings": true,
+  "use_cache": true,
+  "use_mrope": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

neuronxcc-2.18.121.0+9e31e41a/MODULE_2e229618015e416964c5+613edded/compile_flags.json

@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]

neuronxcc-2.18.121.0+9e31e41a/MODULE_2e229618015e416964c5+613edded/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c4d12b378647e94605b87aaf73f6a9bae4ac22ceb71594fd8b7203d4c6129fb6
+size 302392

neuronxcc-2.18.121.0+9e31e41a/MODULE_2e229618015e416964c5+613edded/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0ab2727db7f7ce07daec6a170178e76ec225b4814dd2305b032042cf3cfdf53
+size 1414144

neuronxcc-2.18.121.0+9e31e41a/MODULE_4b31f40da9553c6566ed+613edded/compile_flags.json

@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]

neuronxcc-2.18.121.0+9e31e41a/MODULE_4b31f40da9553c6566ed+613edded/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0addf16990f628a3777cef2364f06a7cc8bcfacd4a38943bf2066c85457a5ed8
+size 384596

neuronxcc-2.18.121.0+9e31e41a/MODULE_4b31f40da9553c6566ed+613edded/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc2eb04bbcea5e3d01459aff14b75c289752943ab1088dff4caae28bfb064ee6
+size 18494464