Synchronizing local compiler cache.

.gitattributes

@@ -4536,3 +4536,5 @@ neuronxcc-2.15.128.0+56dc5a86/MODULE_9528847c3683df34bf8b+39f12043/model.neff fi
 neuronxcc-2.15.128.0+56dc5a86/MODULE_b4c6d9886bd3f12b0085+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.15.128.0+56dc5a86/MODULE_e14bae3ef0ef1bbf7ffb+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.15.128.0+56dc5a86/MODULE_fd943f887217faebbb7b+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_6096e18b12043038e7de+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_f3c227e6e71989ffb769+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.15.128.0+56dc5a86/0_REGISTRY/0.0.25.dev0/inference/llama/ibm/labradorite-13b/b84433eb6423f1445161.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "end_token_id": 32000, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "ibm/labradorite-13b", "checkpoint_revision": "26921cf1d63c9c657e8260d08eaacc14fca43da0", "compiler_type": "neuronx-cc", "compiler_version": "2.15.128.0+56dc5a86", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "pad_token_id": 32001, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32008}

neuronxcc-2.15.128.0+56dc5a86/MODULE_6096e18b12043038e7de+39f12043/compile_flags.json

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

neuronxcc-2.15.128.0+56dc5a86/MODULE_6096e18b12043038e7de+39f12043/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ac9f9ce9c8309e926bb2e96186988a92fb94b6c4219835ad82f2df146c17ff4
+size 437085

neuronxcc-2.15.128.0+56dc5a86/MODULE_6096e18b12043038e7de+39f12043/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f94838cfd2c7dc02ac940cba3468f162cbc1970b9c8b3ead9ec92cc1aac8207
+size 6605824

neuronxcc-2.15.128.0+56dc5a86/MODULE_f3c227e6e71989ffb769+39f12043/compile_flags.json

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

neuronxcc-2.15.128.0+56dc5a86/MODULE_f3c227e6e71989ffb769+39f12043/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c776bed4d1b5a27cbd32cf6789bff1812e460e1874bb6540841a0214854f60e1
+size 560143

neuronxcc-2.15.128.0+56dc5a86/MODULE_f3c227e6e71989ffb769+39f12043/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e945f33306a4e8cee43e141229b7ea3a13363cf752a86fd029e82e0f8860b708
+size 2857984