Hugging Face's logo

aws-neuron
/
optimum-neuron-cache

Model card Files Community
279
dacorvo HF staff commited on
Commit
d143da4
•
1 Parent(s): c633570

Synchronizing local compiler cache.

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +40 -0
  2. neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-0.5B/6d0af0ea85413b9a8bc4.json +1 -0
  3. neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-0.5B/f404b6bcd5e4c946681b.json +1 -0
  4. neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-72B-Instruct/058a352bb7d4c0e8af68.json +1 -0
  5. neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-72B-Instruct/090bc30b9eaa90b56de8.json +1 -0
  6. neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-72B-Instruct/355ef61d7e21c9ca4dfd.json +1 -0
  7. neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-72B-Instruct/5feb3df570ed48ce4ef9.json +1 -0
  8. neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-72B-Instruct/62d648e67c1df9861c81.json +1 -0
  9. neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-72B-Instruct/b0090f8c5a66f58c402e.json +1 -0
  10. neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-Math-7B-Instruct/73abe935a165dd5a8d7d.json +1 -0
  11. neuronxcc-2.15.143.0+e39249ad/MODULE_025073d88f6dec21a13d+39f12043/compile_flags.json +1 -0
  12. neuronxcc-2.15.143.0+e39249ad/MODULE_025073d88f6dec21a13d+39f12043/model.done +0 -0
  13. neuronxcc-2.15.143.0+e39249ad/MODULE_025073d88f6dec21a13d+39f12043/model.hlo_module.pb +3 -0
  14. neuronxcc-2.15.143.0+e39249ad/MODULE_025073d88f6dec21a13d+39f12043/model.neff +0 -0
  15. neuronxcc-2.15.143.0+e39249ad/MODULE_10d128cdbda38f2dd83b+39f12043/compile_flags.json +1 -0
  16. neuronxcc-2.15.143.0+e39249ad/MODULE_10d128cdbda38f2dd83b+39f12043/model.done +0 -0
  17. neuronxcc-2.15.143.0+e39249ad/MODULE_10d128cdbda38f2dd83b+39f12043/model.hlo_module.pb +3 -0
  18. neuronxcc-2.15.143.0+e39249ad/MODULE_10d128cdbda38f2dd83b+39f12043/model.neff +0 -0
  19. neuronxcc-2.15.143.0+e39249ad/MODULE_19508710902bacc178f4+39f12043/compile_flags.json +1 -0
  20. neuronxcc-2.15.143.0+e39249ad/MODULE_19508710902bacc178f4+39f12043/model.done +0 -0
  21. neuronxcc-2.15.143.0+e39249ad/MODULE_19508710902bacc178f4+39f12043/model.hlo_module.pb +3 -0
  22. neuronxcc-2.15.143.0+e39249ad/MODULE_19508710902bacc178f4+39f12043/model.neff +3 -0
  23. neuronxcc-2.15.143.0+e39249ad/MODULE_26a5181917a99c6440c2+39f12043/compile_flags.json +1 -0
  24. neuronxcc-2.15.143.0+e39249ad/MODULE_26a5181917a99c6440c2+39f12043/model.done +0 -0
  25. neuronxcc-2.15.143.0+e39249ad/MODULE_26a5181917a99c6440c2+39f12043/model.hlo_module.pb +3 -0
  26. neuronxcc-2.15.143.0+e39249ad/MODULE_26a5181917a99c6440c2+39f12043/model.neff +3 -0
  27. neuronxcc-2.15.143.0+e39249ad/MODULE_348cd8119f4a5353ee24+39f12043/compile_flags.json +1 -0
  28. neuronxcc-2.15.143.0+e39249ad/MODULE_348cd8119f4a5353ee24+39f12043/model.done +0 -0
  29. neuronxcc-2.15.143.0+e39249ad/MODULE_348cd8119f4a5353ee24+39f12043/model.hlo_module.pb +3 -0
  30. neuronxcc-2.15.143.0+e39249ad/MODULE_348cd8119f4a5353ee24+39f12043/model.neff +3 -0
  31. neuronxcc-2.15.143.0+e39249ad/MODULE_38e952fa2492adf691ff+39f12043/compile_flags.json +1 -0
  32. neuronxcc-2.15.143.0+e39249ad/MODULE_38e952fa2492adf691ff+39f12043/model.done +0 -0
  33. neuronxcc-2.15.143.0+e39249ad/MODULE_38e952fa2492adf691ff+39f12043/model.hlo_module.pb +3 -0
  34. neuronxcc-2.15.143.0+e39249ad/MODULE_38e952fa2492adf691ff+39f12043/model.neff +3 -0
  35. neuronxcc-2.15.143.0+e39249ad/MODULE_445d572321d7551de96d+39f12043/compile_flags.json +1 -0
  36. neuronxcc-2.15.143.0+e39249ad/MODULE_445d572321d7551de96d+39f12043/model.done +0 -0
  37. neuronxcc-2.15.143.0+e39249ad/MODULE_445d572321d7551de96d+39f12043/model.hlo_module.pb +3 -0
  38. neuronxcc-2.15.143.0+e39249ad/MODULE_445d572321d7551de96d+39f12043/model.neff +3 -0
  39. neuronxcc-2.15.143.0+e39249ad/MODULE_4d076c0c56a5768c4069+39f12043/compile_flags.json +1 -0
  40. neuronxcc-2.15.143.0+e39249ad/MODULE_4d076c0c56a5768c4069+39f12043/model.done +0 -0
  41. neuronxcc-2.15.143.0+e39249ad/MODULE_4d076c0c56a5768c4069+39f12043/model.hlo_module.pb +3 -0
  42. neuronxcc-2.15.143.0+e39249ad/MODULE_4d076c0c56a5768c4069+39f12043/model.neff +3 -0
  43. neuronxcc-2.15.143.0+e39249ad/MODULE_5263f71385ed4e05d605+39f12043/compile_flags.json +1 -0
  44. neuronxcc-2.15.143.0+e39249ad/MODULE_5263f71385ed4e05d605+39f12043/model.done +0 -0
  45. neuronxcc-2.15.143.0+e39249ad/MODULE_5263f71385ed4e05d605+39f12043/model.hlo_module.pb +3 -0
  46. neuronxcc-2.15.143.0+e39249ad/MODULE_5263f71385ed4e05d605+39f12043/model.neff +3 -0
  47. neuronxcc-2.15.143.0+e39249ad/MODULE_5425d7325a97f7a4932e+39f12043/compile_flags.json +1 -0
  48. neuronxcc-2.15.143.0+e39249ad/MODULE_5425d7325a97f7a4932e+39f12043/model.done +0 -0
  49. neuronxcc-2.15.143.0+e39249ad/MODULE_5425d7325a97f7a4932e+39f12043/model.hlo_module.pb +3 -0
  50. neuronxcc-2.15.143.0+e39249ad/MODULE_5425d7325a97f7a4932e+39f12043/model.neff +3 -0
.gitattributes CHANGED
@@ -6328,3 +6328,43 @@ neuronxcc-2.15.143.0+e39249ad/MODULE_05d8ecae48e91ee8980d/text_encoder_2/model.n
6328
  neuronxcc-2.15.143.0+e39249ad/MODULE_05d8ecae48e91ee8980d/unet/model.neuron filter=lfs diff=lfs merge=lfs -text
6329
  neuronxcc-2.15.143.0+e39249ad/MODULE_05d8ecae48e91ee8980d/vae_decoder/model.neuron filter=lfs diff=lfs merge=lfs -text
6330
  neuronxcc-2.15.143.0+e39249ad/MODULE_05d8ecae48e91ee8980d/vae_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6328
  neuronxcc-2.15.143.0+e39249ad/MODULE_05d8ecae48e91ee8980d/unet/model.neuron filter=lfs diff=lfs merge=lfs -text
6329
  neuronxcc-2.15.143.0+e39249ad/MODULE_05d8ecae48e91ee8980d/vae_decoder/model.neuron filter=lfs diff=lfs merge=lfs -text
6330
  neuronxcc-2.15.143.0+e39249ad/MODULE_05d8ecae48e91ee8980d/vae_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
6331
+ neuronxcc-2.15.143.0+e39249ad/MODULE_19508710902bacc178f4+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6332
+ neuronxcc-2.15.143.0+e39249ad/MODULE_26a5181917a99c6440c2+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6333
+ neuronxcc-2.15.143.0+e39249ad/MODULE_348cd8119f4a5353ee24+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6334
+ neuronxcc-2.15.143.0+e39249ad/MODULE_38e952fa2492adf691ff+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6335
+ neuronxcc-2.15.143.0+e39249ad/MODULE_445d572321d7551de96d+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6336
+ neuronxcc-2.15.143.0+e39249ad/MODULE_4d076c0c56a5768c4069+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6337
+ neuronxcc-2.15.143.0+e39249ad/MODULE_5263f71385ed4e05d605+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6338
+ neuronxcc-2.15.143.0+e39249ad/MODULE_5425d7325a97f7a4932e+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6339
+ neuronxcc-2.15.143.0+e39249ad/MODULE_577ca4a419d665365921+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6340
+ neuronxcc-2.15.143.0+e39249ad/MODULE_5bba44dbf59b43b3d1f0+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6341
+ neuronxcc-2.15.143.0+e39249ad/MODULE_5c88ac7fb93085af9c2b+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6342
+ neuronxcc-2.15.143.0+e39249ad/MODULE_6158c090f65032e00f74+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6343
+ neuronxcc-2.15.143.0+e39249ad/MODULE_62d7bdfdb721536bb071+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6344
+ neuronxcc-2.15.143.0+e39249ad/MODULE_6402a35ae65f70f2e6b3+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6345
+ neuronxcc-2.15.143.0+e39249ad/MODULE_66477408e960f0ab49ba+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6346
+ neuronxcc-2.15.143.0+e39249ad/MODULE_6e446bc9fb621acf17ec+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6347
+ neuronxcc-2.15.143.0+e39249ad/MODULE_7df64ced3de7c4dfd196+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6348
+ neuronxcc-2.15.143.0+e39249ad/MODULE_7dff79b209c4936eccf6+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6349
+ neuronxcc-2.15.143.0+e39249ad/MODULE_7f495f1e5eab37e231f8+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6350
+ neuronxcc-2.15.143.0+e39249ad/MODULE_822748c7e4759c0698fc+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6351
+ neuronxcc-2.15.143.0+e39249ad/MODULE_8568ef054dcab0a881e6+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6352
+ neuronxcc-2.15.143.0+e39249ad/MODULE_8d073502ce4185d60215+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6353
+ neuronxcc-2.15.143.0+e39249ad/MODULE_8eb7c7f43b1ddb9995d7+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6354
+ neuronxcc-2.15.143.0+e39249ad/MODULE_98773a4d5b572f0b4e19+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6355
+ neuronxcc-2.15.143.0+e39249ad/MODULE_9bf7d4560f134fd504aa+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6356
+ neuronxcc-2.15.143.0+e39249ad/MODULE_9c3a05bace15af5d38e1+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6357
+ neuronxcc-2.15.143.0+e39249ad/MODULE_aa16542849c94822cf60+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6358
+ neuronxcc-2.15.143.0+e39249ad/MODULE_addf3ee88910c0f4f81e+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6359
+ neuronxcc-2.15.143.0+e39249ad/MODULE_b6d1e63d522c57afb5b8+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6360
+ neuronxcc-2.15.143.0+e39249ad/MODULE_bcd777280ee4ee0105a8+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6361
+ neuronxcc-2.15.143.0+e39249ad/MODULE_c59ef31c53c35de18bfa+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6362
+ neuronxcc-2.15.143.0+e39249ad/MODULE_d3c5967570c46d6191a9+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6363
+ neuronxcc-2.15.143.0+e39249ad/MODULE_d71ff4a9a47bb2779232+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6364
+ neuronxcc-2.15.143.0+e39249ad/MODULE_e00e29076f606df4175f+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6365
+ neuronxcc-2.15.143.0+e39249ad/MODULE_e02f71073a5bbd82b90b+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6366
+ neuronxcc-2.15.143.0+e39249ad/MODULE_e63ec966f940f9aac5ed+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6367
+ neuronxcc-2.15.143.0+e39249ad/MODULE_eae3a509248ee092e093+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6368
+ neuronxcc-2.15.143.0+e39249ad/MODULE_f4cf938415821e326a97+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6369
+ neuronxcc-2.15.143.0+e39249ad/MODULE_f7f5761cfc68a7c66eb3+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
6370
+ neuronxcc-2.15.143.0+e39249ad/MODULE_fbc0c1dcbdfb5e889d3e+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-0.5B/6d0af0ea85413b9a8bc4.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Qwen2ForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 151643, "eos_token_id": 151643, "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": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "Qwen/Qwen2.5-0.5B", "checkpoint_revision": "060db6499f32faf8b98477b0a26969ef7d8b9987", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 14, "num_hidden_layers": 24, "num_key_value_heads": 2, "rms_norm_eps": 1e-06, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "use_mrope": false, "use_sliding_window": false, "vocab_size": 151936}
neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-0.5B/f404b6bcd5e4c946681b.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Qwen2ForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 151643, "eos_token_id": 151643, "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": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "Qwen/Qwen2.5-0.5B", "checkpoint_revision": "060db6499f32faf8b98477b0a26969ef7d8b9987", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 14, "num_hidden_layers": 24, "num_key_value_heads": 2, "rms_norm_eps": 1e-06, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "use_mrope": false, "use_sliding_window": false, "vocab_size": 151936}
neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-72B-Instruct/058a352bb7d4c0e8af68.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Qwen2ForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 151643, "eos_token_id": 151645, "hidden_act": "silu", "hidden_size": 8192, "initializer_range": 0.02, "intermediate_size": 29568, "max_position_embeddings": 32768, "max_window_layers": 70, "model_type": "qwen2", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "Qwen/Qwen2.5-72B-Instruct", "checkpoint_revision": "d3d951150c1e5848237cd6a7ad11df4836aee842", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 24, "sequence_length": 32768, "task": "text-generation"}, "num_attention_heads": 64, "num_hidden_layers": 80, "num_key_value_heads": 8, "rms_norm_eps": 1e-06, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "use_sliding_window": false, "vocab_size": 152064}
neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-72B-Instruct/090bc30b9eaa90b56de8.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Qwen2ForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 151643, "eos_token_id": 151645, "hidden_act": "silu", "hidden_size": 8192, "initializer_range": 0.02, "intermediate_size": 29568, "max_position_embeddings": 32768, "max_window_layers": 70, "model_type": "qwen2", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "Qwen/Qwen2.5-72B-Instruct", "checkpoint_revision": "d3d951150c1e5848237cd6a7ad11df4836aee842", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 64, "num_hidden_layers": 80, "num_key_value_heads": 8, "rms_norm_eps": 1e-06, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "use_sliding_window": false, "vocab_size": 152064}
neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-72B-Instruct/355ef61d7e21c9ca4dfd.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Qwen2ForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 151643, "eos_token_id": 151645, "hidden_act": "silu", "hidden_size": 8192, "initializer_range": 0.02, "intermediate_size": 29568, "max_position_embeddings": 32768, "max_window_layers": 70, "model_type": "qwen2", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "Qwen/Qwen2.5-72B-Instruct", "checkpoint_revision": "d3d951150c1e5848237cd6a7ad11df4836aee842", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 24, "sequence_length": 32768, "task": "text-generation"}, "num_attention_heads": 64, "num_hidden_layers": 80, "num_key_value_heads": 8, "rms_norm_eps": 1e-06, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "use_sliding_window": false, "vocab_size": 152064}
neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-72B-Instruct/5feb3df570ed48ce4ef9.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Qwen2ForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 151643, "eos_token_id": 151645, "hidden_act": "silu", "hidden_size": 8192, "initializer_range": 0.02, "intermediate_size": 29568, "max_position_embeddings": 32768, "max_window_layers": 70, "model_type": "qwen2", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "Qwen/Qwen2.5-72B-Instruct", "checkpoint_revision": "d3d951150c1e5848237cd6a7ad11df4836aee842", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 24, "sequence_length": 2048, "task": "text-generation"}, "num_attention_heads": 64, "num_hidden_layers": 80, "num_key_value_heads": 8, "rms_norm_eps": 1e-06, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "use_sliding_window": false, "vocab_size": 152064}
neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-72B-Instruct/62d648e67c1df9861c81.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Qwen2ForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 151643, "eos_token_id": 151645, "hidden_act": "silu", "hidden_size": 8192, "initializer_range": 0.02, "intermediate_size": 29568, "max_position_embeddings": 32768, "max_window_layers": 70, "model_type": "qwen2", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "Qwen/Qwen2.5-72B-Instruct", "checkpoint_revision": "d3d951150c1e5848237cd6a7ad11df4836aee842", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 24, "sequence_length": 3072, "task": "text-generation"}, "num_attention_heads": 64, "num_hidden_layers": 80, "num_key_value_heads": 8, "rms_norm_eps": 1e-06, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "use_sliding_window": false, "vocab_size": 152064}
neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-72B-Instruct/b0090f8c5a66f58c402e.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Qwen2ForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 151643, "eos_token_id": 151645, "hidden_act": "silu", "hidden_size": 8192, "initializer_range": 0.02, "intermediate_size": 29568, "max_position_embeddings": 32768, "max_window_layers": 70, "model_type": "qwen2", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "Qwen/Qwen2.5-72B-Instruct", "checkpoint_revision": "d3d951150c1e5848237cd6a7ad11df4836aee842", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 24, "sequence_length": 3072, "task": "text-generation"}, "num_attention_heads": 64, "num_hidden_layers": 80, "num_key_value_heads": 8, "rms_norm_eps": 1e-06, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "use_sliding_window": false, "vocab_size": 152064}
neuronxcc-2.15.143.0+e39249ad/0_REGISTRY/0.0.27.dev0/inference/qwen2/Qwen/Qwen2.5-Math-7B-Instruct/73abe935a165dd5a8d7d.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Qwen2ForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 151643, "eos_token_id": 151645, "hidden_act": "silu", "hidden_size": 3584, "initializer_range": 0.02, "intermediate_size": 18944, "max_position_embeddings": 4096, "max_window_layers": 28, "model_type": "qwen2", "neuron": {"auto_cast_type": "bf16", "batch_size": 32, "checkpoint_id": "Qwen/Qwen2.5-Math-7B-Instruct", "checkpoint_revision": "ef9926d75ab1d54532f6a30dd5e760355eb9aa4d", "compiler_type": "neuronx-cc", "compiler_version": "2.15.143.0+e39249ad", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 28, "num_hidden_layers": 28, "num_key_value_heads": 4, "rms_norm_eps": 1e-06, "rope_theta": 10000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "use_sliding_window": false, "vocab_size": 152064}
neuronxcc-2.15.143.0+e39249ad/MODULE_025073d88f6dec21a13d+39f12043/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.15.143.0+e39249ad/MODULE_025073d88f6dec21a13d+39f12043/model.done ADDED
File without changes
neuronxcc-2.15.143.0+e39249ad/MODULE_025073d88f6dec21a13d+39f12043/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5f38e8a8f69d59558c745da4a1098dd0e0deba005bf77c0c86dbdfea6d57d2fc
3
+ size 278705
neuronxcc-2.15.143.0+e39249ad/MODULE_025073d88f6dec21a13d+39f12043/model.neff ADDED
Binary file (933 kB). View file
 
neuronxcc-2.15.143.0+e39249ad/MODULE_10d128cdbda38f2dd83b+39f12043/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.15.143.0+e39249ad/MODULE_10d128cdbda38f2dd83b+39f12043/model.done ADDED
File without changes
neuronxcc-2.15.143.0+e39249ad/MODULE_10d128cdbda38f2dd83b+39f12043/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:87def0ac88d7df6d6719764216a6a002a9018794b5ac82665fe9246a4a235da3
3
+ size 278705
neuronxcc-2.15.143.0+e39249ad/MODULE_10d128cdbda38f2dd83b+39f12043/model.neff ADDED
Binary file (871 kB). View file
 
neuronxcc-2.15.143.0+e39249ad/MODULE_19508710902bacc178f4+39f12043/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.15.143.0+e39249ad/MODULE_19508710902bacc178f4+39f12043/model.done ADDED
File without changes
neuronxcc-2.15.143.0+e39249ad/MODULE_19508710902bacc178f4+39f12043/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5042d5b8fa959d76f07f22ee5b51e4f24a4aac185d8c50bc69d5b5353ef72ed8
3
+ size 941713
neuronxcc-2.15.143.0+e39249ad/MODULE_19508710902bacc178f4+39f12043/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:247762813c5117bec51730858c274d65c69391e160996400fa08775ab24e1a52
3
+ size 8193024
neuronxcc-2.15.143.0+e39249ad/MODULE_26a5181917a99c6440c2+39f12043/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.15.143.0+e39249ad/MODULE_26a5181917a99c6440c2+39f12043/model.done ADDED
File without changes
neuronxcc-2.15.143.0+e39249ad/MODULE_26a5181917a99c6440c2+39f12043/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:62c3aa4f33a1233cb66b53cbaaf2f048c2ef4eaf7f6463ed72ce465fd541f904
3
+ size 750092
neuronxcc-2.15.143.0+e39249ad/MODULE_26a5181917a99c6440c2+39f12043/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a572f08d4c78811520f290a62842fd48fb4a48c7aab513acbe9ccbce735fa791
3
+ size 1035264
neuronxcc-2.15.143.0+e39249ad/MODULE_348cd8119f4a5353ee24+39f12043/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.15.143.0+e39249ad/MODULE_348cd8119f4a5353ee24+39f12043/model.done ADDED
File without changes
neuronxcc-2.15.143.0+e39249ad/MODULE_348cd8119f4a5353ee24+39f12043/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d3e114cec205798df7ef944feb91491edd02bb78dd79a897634586c0f9331bda
3
+ size 222784
neuronxcc-2.15.143.0+e39249ad/MODULE_348cd8119f4a5353ee24+39f12043/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:53c9b579b20763b43eec1672b0fbbb7342c9f7d2e9f1e222d8a852da03edddd8
3
+ size 1065984
neuronxcc-2.15.143.0+e39249ad/MODULE_38e952fa2492adf691ff+39f12043/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.15.143.0+e39249ad/MODULE_38e952fa2492adf691ff+39f12043/model.done ADDED
File without changes
neuronxcc-2.15.143.0+e39249ad/MODULE_38e952fa2492adf691ff+39f12043/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a5a832ec2310116b1861f678414c4e5aba7f8fa047b95948989bdf0a8360841d
3
+ size 941713
neuronxcc-2.15.143.0+e39249ad/MODULE_38e952fa2492adf691ff+39f12043/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:53b3f3f60a02b5e71983eca5219119523d804fd87ebe0ae094e50f381cc4007c
3
+ size 8295424
neuronxcc-2.15.143.0+e39249ad/MODULE_445d572321d7551de96d+39f12043/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.15.143.0+e39249ad/MODULE_445d572321d7551de96d+39f12043/model.done ADDED
File without changes
neuronxcc-2.15.143.0+e39249ad/MODULE_445d572321d7551de96d+39f12043/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:457905e0cb64161cbcae2504ca770c64ae2922a38ff31fd31df5eb0220c6963a
3
+ size 750092
neuronxcc-2.15.143.0+e39249ad/MODULE_445d572321d7551de96d+39f12043/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:67ccd6801d7df38b0454752e6a6b96e0a197714123d6b6ade61c7aac511a69c4
3
+ size 9227264
neuronxcc-2.15.143.0+e39249ad/MODULE_4d076c0c56a5768c4069+39f12043/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.15.143.0+e39249ad/MODULE_4d076c0c56a5768c4069+39f12043/model.done ADDED
File without changes
neuronxcc-2.15.143.0+e39249ad/MODULE_4d076c0c56a5768c4069+39f12043/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f72711846367f071f3aad88c865e284b58148912b47602dc2ccbf4d249e5fdbf
3
+ size 750092
neuronxcc-2.15.143.0+e39249ad/MODULE_4d076c0c56a5768c4069+39f12043/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d19c5c00fc234b52105480c9094e53c8612e9fe0f6da06cc015e124d49029705
3
+ size 10589184
neuronxcc-2.15.143.0+e39249ad/MODULE_5263f71385ed4e05d605+39f12043/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.15.143.0+e39249ad/MODULE_5263f71385ed4e05d605+39f12043/model.done ADDED
File without changes
neuronxcc-2.15.143.0+e39249ad/MODULE_5263f71385ed4e05d605+39f12043/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:928c2207ec03af54038c3456aaecda6fe647be9f7cd2d1d2b63be8a5c55069cb
3
+ size 1103969
neuronxcc-2.15.143.0+e39249ad/MODULE_5263f71385ed4e05d605+39f12043/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3b16e17c52195e71e68e8f5711d5ca933f9a414410d4d3b60d79a1916ec3e46c
3
+ size 5479424
neuronxcc-2.15.143.0+e39249ad/MODULE_5425d7325a97f7a4932e+39f12043/compile_flags.json ADDED
@@ -0,0 +1 @@
 
 
1
+ ["--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]
neuronxcc-2.15.143.0+e39249ad/MODULE_5425d7325a97f7a4932e+39f12043/model.done ADDED
File without changes
neuronxcc-2.15.143.0+e39249ad/MODULE_5425d7325a97f7a4932e+39f12043/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fd49b5e7ee2ad050dc5217983f71e0f327853c563ac71e4712822c4a471f1488
3
+ size 941713
neuronxcc-2.15.143.0+e39249ad/MODULE_5425d7325a97f7a4932e+39f12043/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:73f4dca6c5b0d30e15da1037f4cf8723f83549722deee4ed9d86b853e719559c
3
+ size 7957504