Synchronizing local compiler cache.

.gitattributes

@@ -1147,3 +1147,118 @@ neuronxcc-2.15.128.0+56dc5a86/MODULE_fbd5df1f3fbd8961b2cf/model.neuron filter=lf
 neuronxcc-2.15.128.0+56dc5a86/MODULE_fc9ec19cff495325a3e0/model.neuron filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.15.128.0+56dc5a86/MODULE_fd4e98821991f79371a5/model.neuron filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.15.128.0+56dc5a86/MODULE_fe924f54747c457b124f/model.neuron filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_0c9ea6bf6c6bea6f03e0+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_0d1965059c9351f7704e+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_1a8a1dcc5602ff8ddec6+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_1ca67327d6a02ad353a8+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_1cd0565a02c10f144704+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_1effbda15d9f42206720+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_214dc0ee42e04693dbb5+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_22e95c6f73dab6388757+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_2c40bd8a6014a02165ab+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_31af802d5e12d5aa028b+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_3af717c28066870c7216+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_3b383e30294891125e5a+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_40d51447ca85a7759d7f+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_40e160d4eaab44c899f3+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_4ce00b10fe4f29f6dbe1+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_4d56eb1c04b40b3e75de+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_65cdd3c9e39eadcee9b5+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_76ab80e50266df202b78+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_7bdbb30804540e0361a7+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_833e80aecf8c9fa09aca+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_836447b8784a4f817bed+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_85ebea4bab5db49de85f+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_8d41ce0c47e72de277f0+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_9473e53f348276a107d1+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_a6f4ac2633b6b1d204d5+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_b7aecc50b404bc1d53d9+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_c5985f85d97aa97e0809+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_c765d50b5542ffc09176+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_c7a4a1ee44b16d426dc5+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_db2224cc81def0c8b21c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_dcd22245b4a305171833+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_e28ae2c4e84b1d46a704+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_e8582306f63de0ecc01c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_f661f6721bb3f19918f8+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.14.227.0+2d4f85be/MODULE_fa30f35965a9f4fea034+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_00c2465d32b694cdc75a+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_0157f35eb303445e190f+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_02bf38a18ccb8af02b13+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_0b70f795f32d979dd08f+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_0bf56f9a64db27448684+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_0de6c682e356929254c6+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_0eeac286e4dde346cdf4+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_107dec43871d718efda8+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_127dc040363bce8eb922+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_141ebe85aa9f9d0f05ec+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_15684f4b794e72102d51+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_1948c053c0208cb14f17+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_1f352aa63d57587fec70+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_2004627e11b2e5bd46ba+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_216347c7e726ef8144fe+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_21e3dcaee19883fe287e+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_33719efab65aba8971de+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_365ac1cdaf18d8317ae6+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_37b1814c685ac0d82c92+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_3c471449bded502e03a7+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_3f6d05883b6613774923+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_430e78bb7f185201891d+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_458717ec5cd8e550df87+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_45bedf40dd477066a7fa+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_479422f85ce1f5ebd67a+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_5549dfc1cee1cd892362+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_580e488b1c902056533c+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_5ddb9c233d7c07c4b83e+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_5e4c7cfbd9352445f83d+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_5e6dcb3ba87ae7e60d83+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_5ee3ab5ecb033c6bdeae+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_5eefafb0f1f9fc1e59e0+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_6d6c15e58725fa158325+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_703336702854a488dfa4+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_7661d6eadbe802c530f4+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_7719ec8fd597806cec35+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_77f52b3bc8e641473aab+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_79389d88c33464914847+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_7a5881d8f3cfc70fe0d7+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_7c94578e80247ad042b9+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_80adfd88e3f30235a817+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_83867e44a27b65a12481+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_866bbf0d17b9db9b493d+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_87bca61b0e0d50c81990+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_8bdb334c1899e9bed2c5+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_967b0f433d458fa4e23d+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_99a69aadae08714e73ed+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_9ac5df1af559151d7b88+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_9b0eecb5e0b590f9420a+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_a0a5efc77c5084df68b7+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_a3426cf68b987d331b3c+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_a4f3f00fbd1b3e7a58a6+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_aaeee402a55cb205f6f8+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_b01949283e27a04f0e6e+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_b6f31af295bc1fdd0707+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_b720c31dac82c34258cd+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_b73835229038eafd2762+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_bdf4f86a85b75038b651+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_be4df199a79b7377e53a+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_c3e577073bafae1e671b+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_c4a2455c5b25dada65d7+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_c548cc651b8f6e85d9e8+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_c57f19ed333f0e818ff9+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_c65426c729b86309e34a+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_c77c617e0f686c6e9c00+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_cc1727b893df6b3b25c5+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_ce4004e47d7f9259e48d+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_cee1eefa9c3af642bf9b+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_d19b002f79730c2890a1+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_d656e4c29e525a64d500+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_d83898ae6200d4ae727b+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_dfb4e18dbbfd0e05a6e6+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_e9508c7d41b1f29d36fc+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_ec7ce1cb59fb29fdc73b+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_efddfbc496e02d36ff87+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_f10dd3b89a824c879736+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_f34e4ed3c28e63eda19b+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_f4d26174266564b9c206+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_f93de83e0c3ea0bb06cc+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.15.128.0+56dc5a86/MODULE_ff1977202245df118119+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.24/inference/llama/meta-llama/CodeLlama-7b-Instruct-hf/e4d3f9d8d335e8c2ff1f.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 16384, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "meta-llama/CodeLlama-7b-Instruct-hf", "checkpoint_revision": "4ce0c40b2ea823bd1d8f1f3fd5bc8a7e80d749bc", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 1000000, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32016}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/gpt2/gpt2/02b3df5661ef8ff79cc0.json

@@ -0,0 +1 @@
+{"activation_function": "gelu_new", "architectures": ["GPT2LMHeadModel"], "attn_pdrop": 0.1, "bos_token_id": 50256, "embd_pdrop": 0.1, "eos_token_id": 50256, "initializer_range": 0.02, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 1024, "n_embd": 768, "n_head": 12, "n_inner": null, "n_layer": 12, "n_positions": 1024, "neuron": {"auto_cast_type": "f16", "batch_size": 64, "checkpoint_id": "gpt2", "checkpoint_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 24, "sequence_length": 1024, "task": "text-generation"}, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "task_specific_params": {"text-generation": {"do_sample": true, "max_length": 50}}, "use_cache": true, "vocab_size": 50257}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/gpt2/gpt2/4151c6889dcf1584205c.json

@@ -0,0 +1 @@
+{"activation_function": "gelu_new", "architectures": ["GPT2LMHeadModel"], "attn_pdrop": 0.1, "bos_token_id": 50256, "embd_pdrop": 0.1, "eos_token_id": 50256, "initializer_range": 0.02, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 1024, "n_embd": 768, "n_head": 12, "n_inner": null, "n_layer": 12, "n_positions": 1024, "neuron": {"auto_cast_type": "f16", "batch_size": 32, "checkpoint_id": "gpt2", "checkpoint_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 24, "sequence_length": 1024, "task": "text-generation"}, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "task_specific_params": {"text-generation": {"do_sample": true, "max_length": 50}}, "use_cache": true, "vocab_size": 50257}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/gpt2/gpt2/6d5c143cd18bfcd9f4af.json

@@ -0,0 +1 @@
+{"activation_function": "gelu_new", "architectures": ["GPT2LMHeadModel"], "attn_pdrop": 0.1, "bos_token_id": 50256, "embd_pdrop": 0.1, "eos_token_id": 50256, "initializer_range": 0.02, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 1024, "n_embd": 768, "n_head": 12, "n_inner": null, "n_layer": 12, "n_positions": 1024, "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "gpt2", "checkpoint_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 2, "sequence_length": 128, "task": "text-generation"}, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "task_specific_params": {"text-generation": {"do_sample": true, "max_length": 50}}, "use_cache": true, "vocab_size": 50257}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/gpt2/gpt2/874f10ed32bff7260c79.json

@@ -0,0 +1 @@
+{"activation_function": "gelu_new", "architectures": ["GPT2LMHeadModel"], "attn_pdrop": 0.1, "bos_token_id": 50256, "embd_pdrop": 0.1, "eos_token_id": 50256, "initializer_range": 0.02, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 1024, "n_embd": 768, "n_head": 12, "n_inner": null, "n_layer": 12, "n_positions": 1024, "neuron": {"auto_cast_type": "f16", "batch_size": 64, "checkpoint_id": "gpt2", "checkpoint_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 24, "sequence_length": 128, "task": "text-generation"}, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "task_specific_params": {"text-generation": {"do_sample": true, "max_length": 50}}, "use_cache": true, "vocab_size": 50257}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/gpt2/gpt2/befa24d60d5bca012c3b.json

@@ -0,0 +1 @@
+{"activation_function": "gelu_new", "architectures": ["GPT2LMHeadModel"], "attn_pdrop": 0.1, "bos_token_id": 50256, "embd_pdrop": 0.1, "eos_token_id": 50256, "initializer_range": 0.02, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 1024, "n_embd": 768, "n_head": 12, "n_inner": null, "n_layer": 12, "n_positions": 1024, "neuron": {"auto_cast_type": "f16", "batch_size": 1, "checkpoint_id": "gpt2", "checkpoint_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 24, "sequence_length": 1024, "task": "text-generation"}, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "task_specific_params": {"text-generation": {"do_sample": true, "max_length": 50}}, "use_cache": true, "vocab_size": 50257}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/gpt2/hf-internal-testing/tiny-random-gpt2/bba017603eba49cc3a1f.json

@@ -0,0 +1 @@
+{"activation_function": "gelu_new", "attention_probs_dropout_prob": 0.1, "attn_pdrop": 0.1, "bos_token_id": 98, "embd_pdrop": 0.1, "eos_token_id": 98, "gradient_checkpointing": false, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "initializer_range": 0.02, "intermediate_size": 37, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 512, "n_embd": 32, "n_head": 4, "n_inner": null, "n_layer": 5, "n_positions": 512, "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "hf-internal-testing/tiny-random-gpt2", "checkpoint_revision": "71034c5d8bde858ff824298bdedc65515b97d2b9", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 2, "sequence_length": 512, "task": "text-generation"}, "pad_token_id": 98, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "type_vocab_size": 16, "use_cache": true, "vocab_size": 1000}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/gpt2/openai-community/gpt2/180283e8bf01ae3507c0.json

@@ -0,0 +1 @@
+{"activation_function": "gelu_new", "architectures": ["GPT2LMHeadModel"], "attn_pdrop": 0.1, "bos_token_id": 50256, "embd_pdrop": 0.1, "eos_token_id": 50256, "initializer_range": 0.02, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 1024, "n_embd": 768, "n_head": 12, "n_inner": null, "n_layer": 12, "n_positions": 1024, "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "openai-community/gpt2", "checkpoint_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "task_specific_params": {"text-generation": {"do_sample": true, "max_length": 50}}, "use_cache": true, "vocab_size": 50257}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/gpt2/openai-community/gpt2/39a7649c623c1888c776.json

@@ -0,0 +1 @@
+{"activation_function": "gelu_new", "architectures": ["GPT2LMHeadModel"], "attn_pdrop": 0.1, "bos_token_id": 50256, "embd_pdrop": 0.1, "eos_token_id": 50256, "initializer_range": 0.02, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 1024, "n_embd": 768, "n_head": 12, "n_inner": null, "n_layer": 12, "n_positions": 1024, "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "openai-community/gpt2", "checkpoint_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "task_specific_params": {"text-generation": {"do_sample": true, "max_length": 50}}, "use_cache": true, "vocab_size": 50257}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/gpt2/openai-community/gpt2/54e96fc83b91f9ca40c8.json

@@ -0,0 +1 @@
+{"activation_function": "gelu_new", "architectures": ["GPT2LMHeadModel"], "attn_pdrop": 0.1, "bos_token_id": 50256, "embd_pdrop": 0.1, "eos_token_id": 50256, "initializer_range": 0.02, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 1024, "n_embd": 768, "n_head": 12, "n_inner": null, "n_layer": 12, "n_positions": 1024, "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "openai-community/gpt2", "checkpoint_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "task_specific_params": {"text-generation": {"do_sample": true, "max_length": 50}}, "use_cache": true, "vocab_size": 50257}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/llama/meta-llama/Meta-Llama-3-8B-Instruct/54d2c115427eb106b498.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128009, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "meta-llama/Meta-Llama-3-8B-Instruct", "checkpoint_revision": "e1945c40cd546c78e41f1151f4db032b271faeaa", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 24, "sequence_length": 8192, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/llama/meta-llama/Meta-Llama-3-8B-Instruct/55063f94fc7dddda3ac1.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128009, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 8, "checkpoint_id": "meta-llama/Meta-Llama-3-8B-Instruct", "checkpoint_revision": "e1945c40cd546c78e41f1151f4db032b271faeaa", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 24, "sequence_length": 2048, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/llama/meta-llama/Meta-Llama-3-8B-Instruct/832467c16a81d475fbb1.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128009, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "meta-llama/Meta-Llama-3-8B-Instruct", "checkpoint_revision": "e1945c40cd546c78e41f1151f4db032b271faeaa", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/llama/meta-llama/Meta-Llama-3-8B-Instruct/9a95ca98317aed241916.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128009, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 32, "checkpoint_id": "meta-llama/Meta-Llama-3-8B-Instruct", "checkpoint_revision": "e1945c40cd546c78e41f1151f4db032b271faeaa", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 24, "sequence_length": 8192, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/llama/meta-llama/Meta-Llama-3-8B-Instruct/a246f0054be6283dfb7e.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128009, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "meta-llama/Meta-Llama-3-8B-Instruct", "checkpoint_revision": "e1945c40cd546c78e41f1151f4db032b271faeaa", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 24, "sequence_length": 2048, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/llama/meta-llama/Meta-Llama-3-8B-Instruct/b92e78445e785f8ec26d.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128009, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 16, "checkpoint_id": "meta-llama/Meta-Llama-3-8B-Instruct", "checkpoint_revision": "e1945c40cd546c78e41f1151f4db032b271faeaa", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/llama/meta-llama/Meta-Llama-3-8B-Instruct/f1b87f52657555ab6e8e.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128009, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 32, "checkpoint_id": "meta-llama/Meta-Llama-3-8B-Instruct", "checkpoint_revision": "e1945c40cd546c78e41f1151f4db032b271faeaa", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/llama/meta-llama/Meta-Llama-3-8B/2a30c91813b1e7196f95.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128001, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 8, "checkpoint_id": "meta-llama/Meta-Llama-3-8B", "checkpoint_revision": "62bd457b6fe961a42a631306577e622c83876cb6", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/llama/meta-llama/Meta-Llama-3-8B/6434dcc45c7c9439f30f.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128001, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 16, "checkpoint_id": "meta-llama/Meta-Llama-3-8B", "checkpoint_revision": "62bd457b6fe961a42a631306577e622c83876cb6", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/llama/meta-llama/Meta-Llama-3-8B/7e91caedb706bd96bd3a.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128001, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "meta-llama/Meta-Llama-3-8B", "checkpoint_revision": "62bd457b6fe961a42a631306577e622c83876cb6", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/llama/meta-llama/Meta-Llama-3-8B/f22f49b5301c6c906c34.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128001, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "meta-llama/Meta-Llama-3-8B", "checkpoint_revision": "62bd457b6fe961a42a631306577e622c83876cb6", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/llama/meta-llama/Meta-Llama-3-8B/fd9b28c3749a3bbd30c5.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128001, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 32, "checkpoint_id": "meta-llama/Meta-Llama-3-8B", "checkpoint_revision": "62bd457b6fe961a42a631306577e622c83876cb6", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/llama/meta-llama/Meta-Llama-3.1-8B-Instruct/1b591b403ed491d5ab63.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "meta-llama/Meta-Llama-3.1-8B-Instruct", "checkpoint_revision": "5206a32e0bd3067aef1ce90f5528ade7d866253f", "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.14.227.0+2d4f85be/0_REGISTRY/0.0.25.dev0/inference/vit/hf-internal-testing/tiny-random-vit/d41951caf693837ced38.json

@@ -0,0 +1 @@
+{"architectures": ["ViTForImageClassification"], "attention_probs_dropout_prob": 0.1, "encoder_stride": 16, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "hidden_size": 32, "image_size": 30, "initializer_range": 0.02, "intermediate_size": 37, "layer_norm_eps": 1e-12, "model_type": "vit", "neuron": {"auto_cast": null, "auto_cast_type": null, "compiler_type": "neuronx-cc", "compiler_version": "2.14.227.0+2d4f85be", "disable_fallback": false, "disable_fast_relayout": false, "dynamic_batch_size": false, "inline_weights_to_neff": true, "optlevel": "2", "output_attentions": false, "output_hidden_states": false, "static_batch_size": 1, "static_image_size": 30, "static_num_channels": 3, "static_patch_size": 2}, "num_attention_heads": 4, "num_channels": 3, "num_hidden_layers": 5, "patch_size": 2, "qkv_bias": true, "task": "image-classification", "vocab_size": {}}

neuronxcc-2.14.227.0+2d4f85be/MODULE_0c9ea6bf6c6bea6f03e0+2c2d707e/compile_flags.txt

@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none

neuronxcc-2.14.227.0+2d4f85be/MODULE_0c9ea6bf6c6bea6f03e0+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b8e8daf6532c7ead9f96cb22a851fa5f5465806aa4650fed38b0670e63a2872
+size 405658

neuronxcc-2.14.227.0+2d4f85be/MODULE_0c9ea6bf6c6bea6f03e0+2c2d707e/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f987b57583f4739daa220d72544bc18b9c78e20ac7e62b76232d9daad774284
+size 19508224

neuronxcc-2.14.227.0+2d4f85be/MODULE_0d1965059c9351f7704e+2c2d707e/compile_flags.txt

@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none

neuronxcc-2.14.227.0+2d4f85be/MODULE_0d1965059c9351f7704e+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5ed1245ea53f899f203393f819f171f96c1e84311a192afe32ce923432b378aa
+size 263859

neuronxcc-2.14.227.0+2d4f85be/MODULE_0d1965059c9351f7704e+2c2d707e/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8749dd44b2b1c4f9ac28dd73f12f7cdd62d274e45f7031ac573be143f5b159b7
+size 1711104

neuronxcc-2.14.227.0+2d4f85be/MODULE_1140b8ce94a49c64dc0a+2c2d707e/compile_flags.txt

@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none

neuronxcc-2.14.227.0+2d4f85be/MODULE_1140b8ce94a49c64dc0a+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76060ba0acde6101e4b19635759494e2738bcb9f409dac02f9450d1f97c9111b
+size 9926

neuronxcc-2.14.227.0+2d4f85be/MODULE_13824a02aaaf61dc8bcb+2c2d707e/compile_flags.txt

@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none

neuronxcc-2.14.227.0+2d4f85be/MODULE_13824a02aaaf61dc8bcb+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d78f781f32eda7c36b1b6b06560f10237ac5941b291768a6dd1771c96808dae
+size 49626

neuronxcc-2.14.227.0+2d4f85be/MODULE_1410a6ac4741f5c2dc66+2c2d707e/compile_flags.txt

@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none

neuronxcc-2.14.227.0+2d4f85be/MODULE_1410a6ac4741f5c2dc66+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:020d99880f04a1d73c17c2d2e12f8b30c4a34ab852f68872d9938cf298911db4
+size 137155

neuronxcc-2.14.227.0+2d4f85be/MODULE_1817a7b2bbf2fa9752a7+2c2d707e/compile_flags.txt

@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none

neuronxcc-2.14.227.0+2d4f85be/MODULE_1817a7b2bbf2fa9752a7+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7de465b2d26253b2c172b13a940eb18679e418d7ed6a522cd399a9d22ea98276
+size 137155

neuronxcc-2.14.227.0+2d4f85be/MODULE_182575e7fb323521e978+2c2d707e/compile_flags.txt

@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none

neuronxcc-2.14.227.0+2d4f85be/MODULE_182575e7fb323521e978+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ce76c639689087fafee7b701946ea73d7cb9552986aac9d8c3f6ee690bb5e79
+size 137347

neuronxcc-2.14.227.0+2d4f85be/MODULE_1a8a1dcc5602ff8ddec6+2c2d707e/compile_flags.txt

@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none

neuronxcc-2.14.227.0+2d4f85be/MODULE_1a8a1dcc5602ff8ddec6+2c2d707e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9cfdcee8f091f106f06c7a56b329f582076bc1da4c152683076aa230e87fe56c
+size 446653

neuronxcc-2.14.227.0+2d4f85be/MODULE_1a8a1dcc5602ff8ddec6+2c2d707e/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0fa1a6bf375c283388ca78fb71d76f821df577d83c5ea1713b869b181b9eecd7
+size 42302464

neuronxcc-2.14.227.0+2d4f85be/MODULE_1b2f42bbcb326d58aca8+2c2d707e/compile_flags.txt

@@ -0,0 +1 @@
+--model-type=transformer --auto-cast=none