Hugging Face's logo

aws-neuron
/
optimum-neuron-cache

Model card Files Community
252
dacorvo HF staff commited on
Commit
ca3ddd8
•
1 Parent(s): cec1dd1

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 +66 -0
  2. neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/993751574a111b4f623c.json +1 -0
  3. neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/1318765bae79ad4f3364.json +1 -0
  4. neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/6061e791124035a0c3c0.json +1 -0
  5. neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/b3903d2fb025d4712d66.json +1 -0
  6. neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/b55d502618fabb111d87.json +1 -0
  7. neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/ba11ee63b4ca6f2dae4b.json +1 -0
  8. neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/c537409479eecc1148c7.json +1 -0
  9. neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/ef9374b81beb5cd6cd68.json +1 -0
  10. neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/f7445996e04141cff687.json +1 -0
  11. neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/fb363aee1208066639d3.json +1 -0
  12. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/0b9e96edfb00f0f41781.json +1 -0
  13. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/2025960152bb1331bcc8.json +1 -0
  14. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/2311cc196a0f70044562.json +1 -0
  15. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/28dc6684e8ffd1957fcc.json +1 -0
  16. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/2e33fb05af9eda2154f6.json +1 -0
  17. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/4e99ef3d07dc56a0f446.json +1 -0
  18. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/66c4d16781d41c4da7f7.json +1 -0
  19. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/8bf98c2e52fcfe14c5d7.json +1 -0
  20. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/9a378c29a33a95a92f36.json +1 -0
  21. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/baaf1893070cf034975e.json +1 -0
  22. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/da44f10585a29a3d996c.json +1 -0
  23. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/e9e9e0fa42e555b34152.json +1 -0
  24. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/1245bb72b32d75db76f9.json +1 -0
  25. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/64e0d5cd868faca4549f.json +1 -0
  26. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/733bc79383c0cc56537e.json +1 -0
  27. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/ca240a79f952204527fc.json +1 -0
  28. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/eec3579b9122a80133a1.json +1 -0
  29. neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21/inference/mistral/mistralai/Mistral-7B-Instruct-v0.1/0e7141f24a565818c250.json +1 -0
  30. neuronxcc-2.13.66.0+6dfecc895/MODULE_03f901d71e7d6fc084b0+2c2d707e/compile_flags.txt +1 -0
  31. neuronxcc-2.13.66.0+6dfecc895/MODULE_03f901d71e7d6fc084b0+2c2d707e/model.hlo.pb +3 -0
  32. neuronxcc-2.13.66.0+6dfecc895/MODULE_03f901d71e7d6fc084b0+2c2d707e/model.neff +3 -0
  33. neuronxcc-2.13.66.0+6dfecc895/MODULE_061665c139cf2d228da2+2c2d707e/compile_flags.txt +1 -0
  34. neuronxcc-2.13.66.0+6dfecc895/MODULE_061665c139cf2d228da2+2c2d707e/model.hlo.pb +3 -0
  35. neuronxcc-2.13.66.0+6dfecc895/MODULE_061665c139cf2d228da2+2c2d707e/model.neff +3 -0
  36. neuronxcc-2.13.66.0+6dfecc895/MODULE_0661b377a87803f495a3+2c2d707e/compile_flags.txt +1 -0
  37. neuronxcc-2.13.66.0+6dfecc895/MODULE_0661b377a87803f495a3+2c2d707e/model.hlo.pb +3 -0
  38. neuronxcc-2.13.66.0+6dfecc895/MODULE_0661b377a87803f495a3+2c2d707e/model.neff +3 -0
  39. neuronxcc-2.13.66.0+6dfecc895/MODULE_08650f1f182ae04eb005+2c2d707e/compile_flags.txt +1 -0
  40. neuronxcc-2.13.66.0+6dfecc895/MODULE_08650f1f182ae04eb005+2c2d707e/model.hlo.pb +3 -0
  41. neuronxcc-2.13.66.0+6dfecc895/MODULE_08650f1f182ae04eb005+2c2d707e/model.neff +3 -0
  42. neuronxcc-2.13.66.0+6dfecc895/MODULE_08d10b8a53508e0c2751+2c2d707e/compile_flags.txt +1 -0
  43. neuronxcc-2.13.66.0+6dfecc895/MODULE_08d10b8a53508e0c2751+2c2d707e/model.hlo.pb +3 -0
  44. neuronxcc-2.13.66.0+6dfecc895/MODULE_08d10b8a53508e0c2751+2c2d707e/model.neff +3 -0
  45. neuronxcc-2.13.66.0+6dfecc895/MODULE_0d3dfdc13bab640a649c+2c2d707e/compile_flags.txt +1 -0
  46. neuronxcc-2.13.66.0+6dfecc895/MODULE_0d3dfdc13bab640a649c+2c2d707e/model.hlo.pb +3 -0
  47. neuronxcc-2.13.66.0+6dfecc895/MODULE_0d3dfdc13bab640a649c+2c2d707e/model.neff +3 -0
  48. neuronxcc-2.13.66.0+6dfecc895/MODULE_0da3881adb22c468e9dd+2c2d707e/compile_flags.txt +1 -0
  49. neuronxcc-2.13.66.0+6dfecc895/MODULE_0da3881adb22c468e9dd+2c2d707e/model.hlo.pb +3 -0
  50. neuronxcc-2.13.66.0+6dfecc895/MODULE_0da3881adb22c468e9dd+2c2d707e/model.neff +3 -0
.gitattributes CHANGED
@@ -1862,3 +1862,69 @@ neuronxcc-2.13.66.0+6dfecc895/MODULE_1018241485345239569+d41d8cd9/model.neff fil
1862
  neuronxcc-2.13.66.0+6dfecc895/MODULE_1323254031419669677+d41d8cd9/model.neff filter=lfs diff=lfs merge=lfs -text
1863
  neuronxcc-2.13.66.0+6dfecc895/MODULE_15982681761232293785+d41d8cd9/model.neff filter=lfs diff=lfs merge=lfs -text
1864
  neuronxcc-2.13.66.0+6dfecc895/MODULE_17230733582467295311+d41d8cd9/model.neff filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1862
  neuronxcc-2.13.66.0+6dfecc895/MODULE_1323254031419669677+d41d8cd9/model.neff filter=lfs diff=lfs merge=lfs -text
1863
  neuronxcc-2.13.66.0+6dfecc895/MODULE_15982681761232293785+d41d8cd9/model.neff filter=lfs diff=lfs merge=lfs -text
1864
  neuronxcc-2.13.66.0+6dfecc895/MODULE_17230733582467295311+d41d8cd9/model.neff filter=lfs diff=lfs merge=lfs -text
1865
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_03f901d71e7d6fc084b0+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1866
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_061665c139cf2d228da2+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1867
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_0661b377a87803f495a3+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1868
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_08650f1f182ae04eb005+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1869
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_08d10b8a53508e0c2751+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1870
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_0d3dfdc13bab640a649c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1871
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_0da3881adb22c468e9dd+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1872
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_10bb0a24226984cd2e79+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1873
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_1d49e883edce832ceadb+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1874
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_22cbd60a4094506895ea+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1875
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_264ba17751fa266b03f7+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1876
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_26954be27141c8018981+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1877
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_276efc749f5da324ae22+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1878
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_33c177cb394ef699a67e+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1879
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_3605c6371f16a1adcc32+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1880
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_38467f1d12268fd8dbf9+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1881
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_3df7def2c111848ae452+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1882
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_421f42d52033e8896297+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1883
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_438cd8062b5c42fb5257+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1884
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_446e705d692130361857+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1885
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_4a35bbae4b6834a27939+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1886
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_4da51e6c5ebb8979d7ff+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1887
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_584db6317ae7919269d6+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1888
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_5ce07ce0fad817994d3c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1889
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_60618c6fd7ab3a54a2e3+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1890
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_613d40b655171e52e4ed+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1891
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_64422fef1c5a952b303a+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1892
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_66ce923346a83d8d1464+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1893
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_67069d980a3d099431d9+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1894
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_673817fe1ef515a663cd+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1895
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_6c524bf84dc7abaad876+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1896
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_6c9034c50e2b1f7d0a37+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1897
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_769ecd43bffc354a9a49+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1898
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_77d484d6e34b3f944b87+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1899
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_79d75a34ce62de9adc0b+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1900
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_836b3660ff8225436598+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1901
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_885bd504567c00cfe8d0+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1902
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_8d399b75f06f539a3160+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1903
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_9286c5a3c3143356c72d+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1904
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_932c1e8ca18485a5a379+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1905
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_9c0022e28ed4ea762154+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1906
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_9d5297456b55fcbc6010+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1907
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_9dee1ffd8786f06d9857+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1908
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_a3af0aea614065415324+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1909
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_a3f018cbfdf372632114+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1910
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_a4ecedf8f469aabfb0d4+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1911
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_a5580fe3ea33d1e17666+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1912
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_a90ede6c7796ae035869+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1913
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_a95b37443958136b9cbf+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1914
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_a9b055ca8400177aa01a+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1915
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_addd0cc9a49255632d47+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1916
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_b2141128e8c260905609+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1917
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_bdde9cb8b08f1218af4e+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1918
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_c17d7032769c2f5202a8+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1919
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_c296c018a8c3b2a68745+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1920
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_cc7c23b7d4695ed92fc9+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1921
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_cd132d5ec27bc45aa15f+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1922
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_d5669fe05df52025d477+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1923
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_d5a114d6ffc6fa2a086c+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1924
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_e7128b3fdaa681ebacce+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1925
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_eb108b52b7136c6e6793+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1926
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_ebcf4fb0eff00774dce5+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1927
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_ede36215bba4df905d32+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1928
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_fc8603f5cea57b3d97a8+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1929
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_fd536f790fc50a1e7f18+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
1930
+ neuronxcc-2.13.66.0+6dfecc895/MODULE_fe6b7a34619145e26852+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-7b-chat-hf/993751574a111b4f623c.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"vocab_size": 32000, "max_position_embeddings": 4096, "hidden_size": 4096, "intermediate_size": 11008, "num_hidden_layers": 32, "num_attention_heads": 32, "num_key_value_heads": 32, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "pretraining_tp": 1, "use_cache": true, "rope_theta": 10000.0, "rope_scaling": null, "attention_bias": false, "attention_dropout": 0.0, "torch_dtype": "float16", "tie_word_embeddings": false, "architectures": ["LlamaForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 32, "checkpoint_id": "meta-llama/Llama-2-7b-chat-hf", "checkpoint_revision": "92011f62d7604e261f748ec0cfe6329f31193e33", "compiler_type": "neuronx-cc", "compiler_version": "2.12.68.0+4480452af", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}}
neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/1318765bae79ad4f3364.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": 4096, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 10000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"auto_cast_type": "bf16", "batch_size": 16, "checkpoint_id": "mistralai/Mistral-7B-v0.1", "checkpoint_revision": "26bca36bde8333b5d7f72e9ed20ccda6a618af24", "compiler_type": "neuronx-cc", "compiler_version": "2.12.68.0+4480452af", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}}
neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/6061e791124035a0c3c0.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": 4096, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 10000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"auto_cast_type": "bf16", "batch_size": 8, "checkpoint_id": "mistralai/Mistral-7B-v0.1", "checkpoint_revision": "26bca36bde8333b5d7f72e9ed20ccda6a618af24", "compiler_type": "neuronx-cc", "compiler_version": "2.12.68.0+4480452af", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}}
neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/b3903d2fb025d4712d66.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": 4096, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 10000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"task": "text-generation", "batch_size": 8, "num_cores": 8, "auto_cast_type": "bf16", "sequence_length": 4096, "compiler_type": "neuronx-cc", "compiler_version": "2.12.68.0+4480452af", "checkpoint_id": "mistralai/Mistral-7B-v0.1", "checkpoint_revision": "26bca36bde8333b5d7f72e9ed20ccda6a618af24"}}
neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/b55d502618fabb111d87.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": 4096, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 10000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "mistralai/Mistral-7B-v0.1", "checkpoint_revision": "26bca36bde8333b5d7f72e9ed20ccda6a618af24", "compiler_type": "neuronx-cc", "compiler_version": "2.12.68.0+4480452af", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}}
neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/ba11ee63b4ca6f2dae4b.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": 4096, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 10000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"task": "text-generation", "batch_size": 4, "num_cores": 8, "auto_cast_type": "bf16", "sequence_length": 4096, "compiler_type": "neuronx-cc", "compiler_version": "2.12.68.0+4480452af", "checkpoint_id": "mistralai/Mistral-7B-v0.1", "checkpoint_revision": "26bca36bde8333b5d7f72e9ed20ccda6a618af24"}}
neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/c537409479eecc1148c7.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": 4096, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 10000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"auto_cast_type": "bf16", "batch_size": 32, "checkpoint_id": "mistralai/Mistral-7B-v0.1", "checkpoint_revision": "26bca36bde8333b5d7f72e9ed20ccda6a618af24", "compiler_type": "neuronx-cc", "compiler_version": "2.12.68.0+4480452af", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}}
neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/ef9374b81beb5cd6cd68.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": 4096, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 10000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "mistralai/Mistral-7B-v0.1", "checkpoint_revision": "26bca36bde8333b5d7f72e9ed20ccda6a618af24", "compiler_type": "neuronx-cc", "compiler_version": "2.12.68.0+4480452af", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}}
neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/f7445996e04141cff687.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": 4096, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 10000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"task": "text-generation", "batch_size": 16, "num_cores": 8, "auto_cast_type": "bf16", "sequence_length": 4096, "compiler_type": "neuronx-cc", "compiler_version": "2.12.68.0+4480452af", "checkpoint_id": "mistralai/Mistral-7B-v0.1", "checkpoint_revision": "26bca36bde8333b5d7f72e9ed20ccda6a618af24"}}
neuronxcc-2.12.68.0+4480452af/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-v0.1/fb363aee1208066639d3.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"vocab_size": 32000, "max_position_embeddings": 32768, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "sliding_window": 4096, "num_key_value_heads": 8, "hidden_act": "silu", "initializer_range": 0.02, "rms_norm_eps": 1e-05, "use_cache": true, "rope_theta": 10000.0, "attention_dropout": 0.0, "torch_dtype": "bfloat16", "tie_word_embeddings": false, "architectures": ["MistralForCausalLM"], "bos_token_id": 1, "eos_token_id": 2, "model_type": "mistral", "neuron": {"task": "text-generation", "batch_size": 1, "num_cores": 8, "auto_cast_type": "bf16", "sequence_length": 4096, "compiler_type": "neuronx-cc", "compiler_version": "2.12.68.0+4480452af", "checkpoint_id": "mistralai/Mistral-7B-v0.1", "checkpoint_revision": "26bca36bde8333b5d7f72e9ed20ccda6a618af24"}}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/0b9e96edfb00f0f41781.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "29655417e51232f4f2b9b5d3e1418e5a9b04e80e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/2025960152bb1331bcc8.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 8, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "29655417e51232f4f2b9b5d3e1418e5a9b04e80e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 12, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/2311cc196a0f70044562.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 8, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "29655417e51232f4f2b9b5d3e1418e5a9b04e80e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/28dc6684e8ffd1957fcc.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 16, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "29655417e51232f4f2b9b5d3e1418e5a9b04e80e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 12, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/2e33fb05af9eda2154f6.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 8, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "29655417e51232f4f2b9b5d3e1418e5a9b04e80e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/4e99ef3d07dc56a0f446.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 32, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "29655417e51232f4f2b9b5d3e1418e5a9b04e80e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/66c4d16781d41c4da7f7.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "29655417e51232f4f2b9b5d3e1418e5a9b04e80e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/8bf98c2e52fcfe14c5d7.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "29655417e51232f4f2b9b5d3e1418e5a9b04e80e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 12, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/9a378c29a33a95a92f36.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "29655417e51232f4f2b9b5d3e1418e5a9b04e80e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 12, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/baaf1893070cf034975e.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "29655417e51232f4f2b9b5d3e1418e5a9b04e80e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/da44f10585a29a3d996c.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "29655417e51232f4f2b9b5d3e1418e5a9b04e80e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/llama/meta-llama/Llama-2-13b-chat-hf/e9e9e0fa42e555b34152.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 13824, "max_position_embeddings": 4096, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 16, "checkpoint_id": "meta-llama/Llama-2-13b-chat-hf", "checkpoint_revision": "29655417e51232f4f2b9b5d3e1418e5a9b04e80e", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 24, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 40, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/0_REGISTRY/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/1245bb72b32d75db76f9.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": "bf16", "batch_size": 4, "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": 4096, "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/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/64e0d5cd868faca4549f.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": "bf16", "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": 4096, "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/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/733bc79383c0cc56537e.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": "bf16", "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": 4096, "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/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/ca240a79f952204527fc.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": "bf16", "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": 4096, "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/0.0.21.dev0/inference/mistral/mistralai/Mistral-7B-Instruct-v0.2/eec3579b9122a80133a1.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": "bf16", "batch_size": 32, "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": 4096, "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/0.0.21/inference/mistral/mistralai/Mistral-7B-Instruct-v0.1/0e7141f24a565818c250.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": "bf16", "batch_size": 1, "checkpoint_id": "mistralai/Mistral-7B-Instruct-v0.1", "checkpoint_revision": "73068f3702d050a2fd5aa2ca1e612e5036429398", "compiler_type": "neuronx-cc", "compiler_version": "2.13.66.0+6dfecc895", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "rms_norm_eps": 1e-05, "rope_theta": 10000.0, "sliding_window": 4096, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.13.66.0+6dfecc895/MODULE_03f901d71e7d6fc084b0+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_03f901d71e7d6fc084b0+2c2d707e/model.hlo.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1dc92a5b79a0c35630d72fbce4ee7d97d8f697b26b056dac4d71f08d00f250f3
3
+ size 377729
neuronxcc-2.13.66.0+6dfecc895/MODULE_03f901d71e7d6fc084b0+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b31ea96905ca0fdf099f7a42b77e1204897b31bdfc0b3e432cb1013fb975e9cc
3
+ size 2038784
neuronxcc-2.13.66.0+6dfecc895/MODULE_061665c139cf2d228da2+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_061665c139cf2d228da2+2c2d707e/model.hlo.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:630d8da99503ad3710e93aea23f959ec1f22cbf8800f823b361bcf3a0fc6e921
3
+ size 422008
neuronxcc-2.13.66.0+6dfecc895/MODULE_061665c139cf2d228da2+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f622ca628cfa317ae8449a4c9ecc9a84119b8859cc04443873f38ea5de4cf66f
3
+ size 5827584
neuronxcc-2.13.66.0+6dfecc895/MODULE_0661b377a87803f495a3+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_0661b377a87803f495a3+2c2d707e/model.hlo.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:86038fdabee8f4be54390b584458c32ce88b570b85b73bf863d9ba16398f427d
3
+ size 422008
neuronxcc-2.13.66.0+6dfecc895/MODULE_0661b377a87803f495a3+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ebe390ed44c1bb9eeabe3980df2a954db38fc7092d75135e4c2b90693cea63b0
3
+ size 5827584
neuronxcc-2.13.66.0+6dfecc895/MODULE_08650f1f182ae04eb005+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_08650f1f182ae04eb005+2c2d707e/model.hlo.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b965d4a02c872a7c40fb876536b2b7f4a8dd667d5b61f71895e68deae683bfe1
3
+ size 376769
neuronxcc-2.13.66.0+6dfecc895/MODULE_08650f1f182ae04eb005+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e94f7773de16fb236f598b02d0f6dd013b79ab2e82aa298e615b01ceb8a9e4cc
3
+ size 3113984
neuronxcc-2.13.66.0+6dfecc895/MODULE_08d10b8a53508e0c2751+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_08d10b8a53508e0c2751+2c2d707e/model.hlo.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ff4e1b5dcb6d6c3c8ec56b8869dfa5c9d94e4d5d414039964805c7b078fe9ea6
3
+ size 422968
neuronxcc-2.13.66.0+6dfecc895/MODULE_08d10b8a53508e0c2751+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:51604e31669e05750d40b05a29cc6ca8ffed1a9738e0df4fd99fd5b1e17d2497
3
+ size 24771584
neuronxcc-2.13.66.0+6dfecc895/MODULE_0d3dfdc13bab640a649c+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_0d3dfdc13bab640a649c+2c2d707e/model.hlo.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:94ce966826254e120b656b25bb7943857077fae7f9ac9320681baab734c17bdd
3
+ size 478986
neuronxcc-2.13.66.0+6dfecc895/MODULE_0d3dfdc13bab640a649c+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7d1c2cc62895cf41c173ee194e4867181f1ff9c20a3f21f9e599ffa75e13e486
3
+ size 7465984
neuronxcc-2.13.66.0+6dfecc895/MODULE_0da3881adb22c468e9dd+2c2d707e/compile_flags.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ --model-type=transformer --auto-cast=none
neuronxcc-2.13.66.0+6dfecc895/MODULE_0da3881adb22c468e9dd+2c2d707e/model.hlo.pb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1ae9d8cd40ed9b47f09c5d19b24222527a6a775cd1b5e1b047c0d3c9879c5f57
3
+ size 373514
neuronxcc-2.13.66.0+6dfecc895/MODULE_0da3881adb22c468e9dd+2c2d707e/model.neff ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:597490ec7d69ace816dd1f1c54ef318ce6b0349eb9eb6b2a53e3c614203c0c68
3
+ size 5858304