Hugging Face's logo

aws-neuron
/
optimum-neuron-cache

Model card Files Community
240
jburtoft commited on
Commit
2ef1a4f
•
1 Parent(s): 70031e1

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 +65 -0
  2. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/17b67fc1a6b8c8e05e23.json +1 -0
  3. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/3d95490ec3b2261788f9.json +1 -0
  4. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/4c806cf17d6a70988558.json +1 -0
  5. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/56e53970323e67087715.json +1 -0
  6. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/5a159d4e29586c7f7fd8.json +1 -0
  7. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/68af53f23d7915fe2bfb.json +1 -0
  8. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/98be56287ed3d7b07c60.json +1 -0
  9. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/aa5d3fe6d2344794f1ab.json +1 -0
  10. neuronxcc-2.13.66.0+6dfecc895/MODULE_05132de1604630122c7d+2c2d707e/compile_flags.txt +1 -0
  11. neuronxcc-2.13.66.0+6dfecc895/MODULE_05132de1604630122c7d+2c2d707e/model.hlo_module.pb +3 -0
  12. neuronxcc-2.13.66.0+6dfecc895/MODULE_05132de1604630122c7d+2c2d707e/model.neff +3 -0
  13. neuronxcc-2.13.66.0+6dfecc895/MODULE_085c37d97eaa63d707cf+2c2d707e/compile_flags.txt +1 -0
  14. neuronxcc-2.13.66.0+6dfecc895/MODULE_085c37d97eaa63d707cf+2c2d707e/model.hlo_module.pb +3 -0
  15. neuronxcc-2.13.66.0+6dfecc895/MODULE_085c37d97eaa63d707cf+2c2d707e/model.neff +3 -0
  16. neuronxcc-2.13.66.0+6dfecc895/MODULE_0edb4d25c5726ca11b17+2c2d707e/compile_flags.txt +1 -0
  17. neuronxcc-2.13.66.0+6dfecc895/MODULE_0edb4d25c5726ca11b17+2c2d707e/model.hlo_module.pb +3 -0
  18. neuronxcc-2.13.66.0+6dfecc895/MODULE_0edb4d25c5726ca11b17+2c2d707e/model.neff +3 -0
  19. neuronxcc-2.13.66.0+6dfecc895/MODULE_0eea048fb87feec0ff22+2c2d707e/compile_flags.txt +1 -0
  20. neuronxcc-2.13.66.0+6dfecc895/MODULE_0eea048fb87feec0ff22+2c2d707e/model.hlo_module.pb +3 -0
  21. neuronxcc-2.13.66.0+6dfecc895/MODULE_0eea048fb87feec0ff22+2c2d707e/model.neff +3 -0
  22. neuronxcc-2.13.66.0+6dfecc895/MODULE_16a40e5a106fae2fac45+2c2d707e/compile_flags.txt +1 -0
  23. neuronxcc-2.13.66.0+6dfecc895/MODULE_16a40e5a106fae2fac45+2c2d707e/model.hlo_module.pb +3 -0
  24. neuronxcc-2.13.66.0+6dfecc895/MODULE_16a40e5a106fae2fac45+2c2d707e/model.neff +3 -0
  25. neuronxcc-2.13.66.0+6dfecc895/MODULE_28e33d89a8dd20024379+2c2d707e/compile_flags.txt +1 -0
  26. neuronxcc-2.13.66.0+6dfecc895/MODULE_28e33d89a8dd20024379+2c2d707e/model.hlo_module.pb +3 -0
  27. neuronxcc-2.13.66.0+6dfecc895/MODULE_28e33d89a8dd20024379+2c2d707e/model.neff +3 -0
  28. neuronxcc-2.13.66.0+6dfecc895/MODULE_29d670fbb82e31bed44f+2c2d707e/compile_flags.txt +1 -0
  29. neuronxcc-2.13.66.0+6dfecc895/MODULE_29d670fbb82e31bed44f+2c2d707e/model.hlo_module.pb +3 -0
  30. neuronxcc-2.13.66.0+6dfecc895/MODULE_29d670fbb82e31bed44f+2c2d707e/model.neff +3 -0
  31. neuronxcc-2.13.66.0+6dfecc895/MODULE_2a22846ebbeaedfcbce0+2c2d707e/compile_flags.txt +1 -0
  32. neuronxcc-2.13.66.0+6dfecc895/MODULE_2a22846ebbeaedfcbce0+2c2d707e/model.hlo_module.pb +3 -0
  33. neuronxcc-2.13.66.0+6dfecc895/MODULE_2a22846ebbeaedfcbce0+2c2d707e/model.neff +3 -0
  34. neuronxcc-2.13.66.0+6dfecc895/MODULE_2d7acb113663e8fd12c4+2c2d707e/compile_flags.txt +1 -0
  35. neuronxcc-2.13.66.0+6dfecc895/MODULE_2d7acb113663e8fd12c4+2c2d707e/model.hlo_module.pb +3 -0
  36. neuronxcc-2.13.66.0+6dfecc895/MODULE_2d7acb113663e8fd12c4+2c2d707e/model.neff +3 -0
  37. neuronxcc-2.13.66.0+6dfecc895/MODULE_2e3824608a119ab96853+2c2d707e/compile_flags.txt +1 -0
  38. neuronxcc-2.13.66.0+6dfecc895/MODULE_2e3824608a119ab96853+2c2d707e/model.hlo_module.pb +3 -0
  39. neuronxcc-2.13.66.0+6dfecc895/MODULE_2e3824608a119ab96853+2c2d707e/model.neff +3 -0
  40. neuronxcc-2.13.66.0+6dfecc895/MODULE_32b831115f4875451ac7+2c2d707e/compile_flags.txt +1 -0
  41. neuronxcc-2.13.66.0+6dfecc895/MODULE_32b831115f4875451ac7+2c2d707e/model.hlo_module.pb +3 -0
  42. neuronxcc-2.13.66.0+6dfecc895/MODULE_32b831115f4875451ac7+2c2d707e/model.neff +3 -0
  43. neuronxcc-2.13.66.0+6dfecc895/MODULE_32c72eb454ede22bc202+2c2d707e/compile_flags.txt +1 -0
  44. neuronxcc-2.13.66.0+6dfecc895/MODULE_32c72eb454ede22bc202+2c2d707e/model.hlo_module.pb +3 -0
  45. neuronxcc-2.13.66.0+6dfecc895/MODULE_32c72eb454ede22bc202+2c2d707e/model.neff +3 -0
  46. neuronxcc-2.13.66.0+6dfecc895/MODULE_33cdfad3be8fab97176f+2c2d707e/compile_flags.txt +1 -0
  47. neuronxcc-2.13.66.0+6dfecc895/MODULE_33cdfad3be8fab97176f+2c2d707e/model.hlo_module.pb +3 -0
  48. neuronxcc-2.13.66.0+6dfecc895/MODULE_33cdfad3be8fab97176f+2c2d707e/model.neff +3 -0
  49. neuronxcc-2.13.66.0+6dfecc895/MODULE_33dba9f408e8826f48b5+2c2d707e/compile_flags.txt +1 -0
  50. neuronxcc-2.13.66.0+6dfecc895/MODULE_33dba9f408e8826f48b5+2c2d707e/model.hlo_module.pb +3 -0
.gitattributes CHANGED
@@ -3005,3 +3005,68 @@ neuronxcc-2.13.66.0+6dfecc895/MODULE_42df9e1a521f5732af09+2c2d707e/model.neff fi
3005
  neuronxcc-2.13.66.0+6dfecc895/MODULE_4cc7cdbf96054f333830+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3006
  neuronxcc-2.13.66.0+6dfecc895/MODULE_80b37e060185910ca232+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3007
  neuronxcc-2.13.66.0+6dfecc895/MODULE_e805a82c8d8a8e7ebfe1+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3005
  neuronxcc-2.13.66.0+6dfecc895/MODULE_4cc7cdbf96054f333830+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3006
  neuronxcc-2.13.66.0+6dfecc895/MODULE_80b37e060185910ca232+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3007
  neuronxcc-2.13.66.0+6dfecc895/MODULE_e805a82c8d8a8e7ebfe1+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3008
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_05132de1604630122c7d+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3009
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_085c37d97eaa63d707cf+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3010
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_0edb4d25c5726ca11b17+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3011
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_0eea048fb87feec0ff22+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3012
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_16a40e5a106fae2fac45+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3013
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_28e33d89a8dd20024379+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3014
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_29d670fbb82e31bed44f+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3015
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_2a22846ebbeaedfcbce0+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3016
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_2d7acb113663e8fd12c4+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3017
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_2e3824608a119ab96853+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3018
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_32b831115f4875451ac7+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3019
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_32c72eb454ede22bc202+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3020
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_33cdfad3be8fab97176f+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3021
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_33dba9f408e8826f48b5+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3022
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_39a977504815ee41c24e+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3023
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_3cfa5ab4c941c6e50f57+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3024
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_3ece8a4c15579f84a689+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3025
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_490043214bf448b8d65c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3026
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_4ddb36823d13429fb206+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3027
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_522a6c2d4f6e5dcb03de+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3028
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_53c7a5c93c3973d77592+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3029
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_5c77eea8a0c44fd18041+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3030
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_613b06d868b38d73fb7a+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3031
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_66a9354c32fca0d8b4dd+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3032
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_686b8cf2bca296f8cc4c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3033
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_694a4a01f843ea08e88e+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3034
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_6b8df1b192490a788a39+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3035
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_6d2fb5743cd05c88feb4+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3036
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_6f54b5324d82997e7b0b+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3037
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_72a03640c5fac2fe8179+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3038
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_781a464d692411330cf3+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3039
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_7d8cbe48b805f6ed66eb+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3040
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_7e42264f0010959abfa8+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3041
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_7e8c409d0b348a61dee6+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3042
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_7f0be0f9e0e73b16b357+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3043
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_87be25dd6640429f24c1+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3044
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_87f0405fd368dc5f2c58+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3045
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_8b615596d2d2e816d4b2+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3046
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_9d878bfcdfe75f201f55+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3047
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_9db639bc10fbddf5832c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3048
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_9df506fee640eb399e8b+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3049
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_aed55d3f7e18d6ac84eb+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3050
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_aff5ac11b28982b8542f+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3051
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_b249e5c7673576c82bf4+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3052
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_bbd56f7a3565a917a3b0+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3053
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_bf6c02083680b0beb5d0+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3054
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_c02a123322346fde75c1+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3055
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_c092d6f9668781eb4582+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3056
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_c8a53af82c1d239874f7+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3057
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_cd858d5da94c433cf8b2+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3058
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_ce700bb92bde210702e9+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3059
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_cec9ea7d1f9a8c4dd516+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3060
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_d2ab568c40ef2e5ff5ee+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3061
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_d2b5a46d19324224114d+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3062
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_d3754a15c60143e13874+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3063
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_d3bd5e0986a03bd7af7a+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3064
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_d5b30177c7efd37d4162+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3065
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_dd35b812dfc79ab6fb98+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3066
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_dfb1876eee13b2984e61+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3067
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_e020139ba890ac28eed6+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3068
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_e317c0c7525f7b05dc9e+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3069
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_e7909ceae4eb5f438d21+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3070
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_f3d2562749cdbfdea433+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3071
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_f3ec010b87c78f434d00+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
3072
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_f599ed1b647e181abf84+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/17b67fc1a6b8c8e05e23.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 32768, "model_type": "mistral", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.2", "checkpoint_revision": "41b61a33a2483885c981aa79e0df6b32407ed873", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 12, "sequence_length": 2048, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "rms_norm_eps": 1e-05, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/3d95490ec3b2261788f9.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 32768, "model_type": "mistral", "neuron": {"auto_cast_type": "fp16", "batch_size": 16, "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.2", "checkpoint_revision": "41b61a33a2483885c981aa79e0df6b32407ed873", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 12, "sequence_length": 2048, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "rms_norm_eps": 1e-05, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/4c806cf17d6a70988558.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 32768, "model_type": "mistral", "neuron": {"auto_cast_type": "fp16", "batch_size": 16, "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.2", "checkpoint_revision": "41b61a33a2483885c981aa79e0df6b32407ed873", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 2048, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "rms_norm_eps": 1e-05, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/56e53970323e67087715.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 32768, "model_type": "mistral", "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.2", "checkpoint_revision": "41b61a33a2483885c981aa79e0df6b32407ed873", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 12, "sequence_length": 2048, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "rms_norm_eps": 1e-05, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/5a159d4e29586c7f7fd8.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 32768, "model_type": "mistral", "neuron": {"auto_cast_type": "fp16", "batch_size": 8, "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.2", "checkpoint_revision": "41b61a33a2483885c981aa79e0df6b32407ed873", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 2048, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "rms_norm_eps": 1e-05, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/68af53f23d7915fe2bfb.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 32768, "model_type": "mistral", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.2", "checkpoint_revision": "41b61a33a2483885c981aa79e0df6b32407ed873", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 2048, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "rms_norm_eps": 1e-05, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/98be56287ed3d7b07c60.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 32768, "model_type": "mistral", "neuron": {"auto_cast_type": "fp16", "batch_size": 8, "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.2", "checkpoint_revision": "41b61a33a2483885c981aa79e0df6b32407ed873", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 12, "sequence_length": 2048, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "rms_norm_eps": 1e-05, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.22/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/aa5d3fe6d2344794f1ab.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 32768, "model_type": "mistral", "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.2", "checkpoint_revision": "41b61a33a2483885c981aa79e0df6b32407ed873", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 2048, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "rms_norm_eps": 1e-05, "rope_theta": 1000000.0, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/MODULE_05132de1604630122c7d+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_05132de1604630122c7d+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fc2f5d7030e169552867cf6fede4aaaabea74a1e9c116135742a2bbcd360038e
3
+ size 297626
neuronxcc-2.13.66.0+6dfecc895/MODULE_05132de1604630122c7d+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3b2dfbfac01be7554eeb0cd6639cd077baa560eb57e98a978fa1b98bdf466be4
3
+ size 8807424
neuronxcc-2.13.66.0+6dfecc895/MODULE_085c37d97eaa63d707cf+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_085c37d97eaa63d707cf+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:10a2adaae9d6a7b9beae1b8217679e4751995ca9b8e8be8689dc1f8097e61cf5
3
+ size 294331
neuronxcc-2.13.66.0+6dfecc895/MODULE_085c37d97eaa63d707cf+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:107a2fbde15d1cd9604812267916e683405cb1c9cf328fe38bd2d2303310d165
3
+ size 2345984
neuronxcc-2.13.66.0+6dfecc895/MODULE_0edb4d25c5726ca11b17+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_0edb4d25c5726ca11b17+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ec1ff99ab9e409b50950f1f49af8710d4a2887cdd750b4197a1290dd75c1b5bf
3
+ size 300985
neuronxcc-2.13.66.0+6dfecc895/MODULE_0edb4d25c5726ca11b17+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:569e45570b1712089aed3fa1d2f74a7085c04b9e020122fae5541d92729a9c20
3
+ size 7937024
neuronxcc-2.13.66.0+6dfecc895/MODULE_0eea048fb87feec0ff22+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_0eea048fb87feec0ff22+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:489c420d776f38261c9ac4a8a1742fd7c412cc9d36cddc618584443a8e2f012f
3
+ size 300985
neuronxcc-2.13.66.0+6dfecc895/MODULE_0eea048fb87feec0ff22+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4afcb65339681c9a90cdcc58d6dc224946fa4f28395d75173ff6657bd7cbc8f4
3
+ size 7906304
neuronxcc-2.13.66.0+6dfecc895/MODULE_16a40e5a106fae2fac45+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_16a40e5a106fae2fac45+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:536dfdab43b3bda027c43be640400a76e6fc61f1f89fdaa9ff29d27d3647b0e0
3
+ size 294331
neuronxcc-2.13.66.0+6dfecc895/MODULE_16a40e5a106fae2fac45+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:39b9c39df74e9e33f597eb6a59e412dda0a3d091df39706d30bdb612d1165e2c
3
+ size 2499584
neuronxcc-2.13.66.0+6dfecc895/MODULE_28e33d89a8dd20024379+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_28e33d89a8dd20024379+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cfa5133c61036815375c32cc278fe748b739ddef52ef55e1eb581e0c8bb1ee8d
3
+ size 399416
neuronxcc-2.13.66.0+6dfecc895/MODULE_28e33d89a8dd20024379+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:47cfe3ec9677d2efb000950af8887801138091c2b7363ca671614053251eab7f
3
+ size 3124224
neuronxcc-2.13.66.0+6dfecc895/MODULE_29d670fbb82e31bed44f+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_29d670fbb82e31bed44f+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c3786d1adf39a86b24a9b9f59a57d2b43ede4e8f86f3b1ef03781049515ac6c0
3
+ size 297626
neuronxcc-2.13.66.0+6dfecc895/MODULE_29d670fbb82e31bed44f+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:47f7b02df06e08f8e953fa7a2445711a7641bfccf763e9a5df3ae59f0b8f55c1
3
+ size 39896064
neuronxcc-2.13.66.0+6dfecc895/MODULE_2a22846ebbeaedfcbce0+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_2a22846ebbeaedfcbce0+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b0854d3a198ebbb5cb6ce1bb9a98128ee4880dda0167eb0ef597bc3cd0494322
3
+ size 297626
neuronxcc-2.13.66.0+6dfecc895/MODULE_2a22846ebbeaedfcbce0+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a8ba397c7e6418beface90b96184e28498500355f1dd73e17c1c43bdc7789601
3
+ size 9933824
neuronxcc-2.13.66.0+6dfecc895/MODULE_2d7acb113663e8fd12c4+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_2d7acb113663e8fd12c4+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c50cf92e4a33c31c0d6bd353fafca3226767628bcc6abe6b1b3e1dd7ec74a0d7
3
+ size 337804
neuronxcc-2.13.66.0+6dfecc895/MODULE_2d7acb113663e8fd12c4+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1c6c3f14458e5acfed043cdf3842a2bfeb3b60874af4d599a772e94547618edd
3
+ size 11131904
neuronxcc-2.13.66.0+6dfecc895/MODULE_2e3824608a119ab96853+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_2e3824608a119ab96853+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0110e68ac5cdae79f1e85b753f9f2b566d1c764c877680db22a2655a12ecfdfc
3
+ size 301369
neuronxcc-2.13.66.0+6dfecc895/MODULE_2e3824608a119ab96853+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:04822e97d9ab8bfe3db6ccac8bbe7404509994d34bd2d08809b2d777f85cb38d
3
+ size 2315264
neuronxcc-2.13.66.0+6dfecc895/MODULE_32b831115f4875451ac7+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_32b831115f4875451ac7+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2ccb4f38bc56760a880fee7a60ac7d49303db425d91583354630cd93f00f9438
3
+ size 293947
neuronxcc-2.13.66.0+6dfecc895/MODULE_32b831115f4875451ac7+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:157b500b841db41d40809c8e362dbb729f1f0a180f52bf8910c373c5ff84206c
3
+ size 8100864
neuronxcc-2.13.66.0+6dfecc895/MODULE_32c72eb454ede22bc202+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_32c72eb454ede22bc202+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:711ba72f20d6d9d4f8018030a7ee7d017a407311ee85ffad506e16f71956f178
3
+ size 294331
neuronxcc-2.13.66.0+6dfecc895/MODULE_32c72eb454ede22bc202+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:05c277ed6a6bd71e894674b7c5c15eab0b5020517b0eff4d77516ed547b469dc
3
+ size 2335744
neuronxcc-2.13.66.0+6dfecc895/MODULE_33cdfad3be8fab97176f+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_33cdfad3be8fab97176f+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:49af950a10cbafa0622c5b72e47ff147f571a9fc0d6d45b4d9118630e940cfec
3
+ size 301369
neuronxcc-2.13.66.0+6dfecc895/MODULE_33cdfad3be8fab97176f+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:88a5c22080b2f0d90ed805c58ff50accdd209c49326638cedb05f096ce690fc7
3
+ size 2294784
neuronxcc-2.13.66.0+6dfecc895/MODULE_33dba9f408e8826f48b5+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_33dba9f408e8826f48b5+2c2d707e/model.hlo_module.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1bd8aabc96445498db9d332ca0900688066f8f5447131defc0c262d5d4c5c749
3
+ size 301369