Hugging Face's logo

dacorvo
/
optimum-neuron-cache

Model card Files
xet
 Community
dacorvo HF Staff commited on
Commit
5860b03
·
verified ·
1 Parent(s): d7f5b1f

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 +269 -0
  2. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/gpt2/gpt2/780a42b2f8414c10d661.json +1 -0
  3. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/granite/ibm-granite/granite-3.1-2b-instruct/0211503669c9e68fbbde.json +1 -0
  4. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/granite/ibm-granite/granite-3.1-2b-instruct/7cf1a953f534c2649a7d.json +1 -0
  5. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/NousResearch/Hermes-2-Theta-Llama-3-8B/7bae97d51948b959db6b.json +1 -0
  6. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/meta-llama/Llama-3.1-8B-Instruct/25395edbfe4cb40ed4e9.json +1 -0
  7. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/meta-llama/Llama-3.2-1B-Instruct/7a87208395f3db5ba7be.json +1 -0
  8. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/meta-llama/Llama-3.2-1B-Instruct/8cd73c74d69aa6440798.json +1 -0
  9. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/meta-llama/Llama-3.2-1B-Instruct/ebbf84c3779811da8d2c.json +1 -0
  10. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/meta-llama/Meta-Llama-3.1-8B-Instruct/3b6b0065e24ca9d2eeb0.json +1 -0
  11. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/princeton-nlp/Sheared-LLaMA-1.3B/8f3a6f200be91cc32a57.json +1 -0
  12. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/mistral/optimum/mistral-1.1b-testing/6668357e5d6bb2f40fe7.json +1 -0
  13. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/mixtral/dacorvo/Mixtral-tiny/f574775a377747e5836b.json +1 -0
  14. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-3-medium-4k-instruct/ae83aa0fea9a9d0315d5.json +1 -0
  15. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-3-mini-4k-instruct/13bb5f11971eace7eeed.json +1 -0
  16. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-3-mini-4k-instruct/29904a0c0edb4679ee76.json +1 -0
  17. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-3-mini-4k-instruct/9a957125ff99f484a66e.json +1 -0
  18. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-3-mini-4k-instruct/d7892042a1da5eaa0020.json +1 -0
  19. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-4-mini-instruct/6ff6f9ae27117fd24bd7.json +1 -0
  20. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/42644c76f78cfd9108b4.json +1 -0
  21. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/452310f61d209c411ec5.json +1 -0
  22. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/6faff6d0347b9b3197ed.json +1 -0
  23. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/aefe5e89e89e5ba40a59.json +1 -0
  24. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/b05887d807c6f53298d2.json +1 -0
  25. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/c488c1faf1a22992a2ea.json +1 -0
  26. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/e21879dbd09a21fb7a2e.json +1 -0
  27. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/f6232f240b5b98678a51.json +1 -0
  28. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/yujiepan/phi-4-tiny-random/33891c5b973ec2bfb226.json +1 -0
  29. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/yujiepan/phi-4-tiny-random/48d7f1c1e412ea84b60e.json +1 -0
  30. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/yujiepan/phi-4-tiny-random/a37d77e276a4a0f5c4e8.json +1 -0
  31. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/qwen2/Qwen/Qwen2.5-0.5B/08b82ea5dab18178ad7f.json +1 -0
  32. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/qwen2/Qwen/Qwen2.5-Math-1.5B/a5b11f2dd182bd813171.json +1 -0
  33. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/058e3ead348125e7808b.json +1 -0
  34. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/6bf9ef24cedd2d181630.json +1 -0
  35. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/7a4c62323b876a6ad082.json +1 -0
  36. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/gpt2/gpt2/780a42b2f8414c10d661.json +1 -0
  37. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/gpt2/hf-internal-testing/tiny-random-gpt2/3033d959070fc853146f.json +1 -0
  38. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/gpt2/hf-internal-testing/tiny-random-gpt2/46dc72bec418e48dc0b2.json +1 -0
  39. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/gpt2/hf-internal-testing/tiny-random-gpt2/fdfe8a25cf9f41629cb9.json +1 -0
  40. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/granite/ibm-granite/granite-3.1-2b-instruct/bf71032506a251068193.json +1 -0
  41. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/llama/Maykeye/TinyLLama-v0/f71b7fdfb0dfed1fac13.json +1 -0
  42. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/llama/NousResearch/Hermes-2-Theta-Llama-3-8B/7bae97d51948b959db6b.json +1 -0
  43. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/llama/llamafactory/tiny-random-Llama-3/2e8cb42c92e334dfbaa9.json +1 -0
  44. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/llama/llamafactory/tiny-random-Llama-3/5433fffb45752137c8de.json +1 -0
  45. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/llama/llamafactory/tiny-random-Llama-3/8c48d1da1325809763e8.json +1 -0
  46. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/llama/unsloth/Llama-3.2-1B-Instruct/49d61e18927b113552d1.json +1 -0
  47. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/3afe2ef31d28f1e456be.json +1 -0
  48. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/aa9d07318c2ee53a79c9.json +1 -0
  49. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/ef87caaa517e593a73ba.json +1 -0
  50. neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/mistral/optimum/mistral-1.1b-testing/6668357e5d6bb2f40fe7.json +1 -0
.gitattributes CHANGED
@@ -327,3 +327,272 @@ neuronxcc-2.12.68.0+4480452af/MODULE_688375d84026af8f7c57+2c2d707e/model.neff fi
327
  neuronxcc-2.12.68.0+4480452af/MODULE_7714327f48e85f09fd03+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
328
  neuronxcc-2.12.68.0+4480452af/MODULE_afb3b0eeaa2511ec5efd+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
329
  neuronxcc-2.12.68.0+4480452af/MODULE_e053a935644ddb80b9eb+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
327
  neuronxcc-2.12.68.0+4480452af/MODULE_7714327f48e85f09fd03+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
328
  neuronxcc-2.12.68.0+4480452af/MODULE_afb3b0eeaa2511ec5efd+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
329
  neuronxcc-2.12.68.0+4480452af/MODULE_e053a935644ddb80b9eb+2c2d707e/model.neff filter=lfs diff=lfs merge=lfs -text
330
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_00804590004fa5eab7c9+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
331
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_008512cdad534106824b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
332
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_00c7ea8a2ca42389a896+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
333
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_01c445025d99d71be2b3+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
334
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_0295f5e523e3ea7df756+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
335
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_032f74178031f5ed3c74+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
336
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_05b0003a35326e3b2f57+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
337
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_068ef074748718999f13+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
338
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_06ca6d62f0361998714e+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
339
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_0977746a68f18f7b9228+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
340
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_09d665c9864fd54d66b2+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
341
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_0a88901e8c98f54e4c10+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
342
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_0a8fcf8ddb59d8730a24+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
343
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_0b137632bac8c4726d4f/decoder/model.neuron filter=lfs diff=lfs merge=lfs -text
344
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_0b137632bac8c4726d4f/encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
345
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_0c3610ac3206c8024aae+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
346
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_0d5e2c0e852741a3814c+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
347
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_1090c91454701eac92a9+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
348
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_10bd954bb3f712b9cf93+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
349
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_11648890582807925826+3cc9a3cb/model.neff filter=lfs diff=lfs merge=lfs -text
350
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_12186797013122181295+3cc9a3cb/model.neff filter=lfs diff=lfs merge=lfs -text
351
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_13309912121921126684+3cc9a3cb/model.neff filter=lfs diff=lfs merge=lfs -text
352
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_13458a001a4a6d6c5895+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
353
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_1505680390210029555+3cc9a3cb/model.neff filter=lfs diff=lfs merge=lfs -text
354
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_151e013069d6b102df91+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
355
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_15431189573610337239+3cc9a3cb/model.neff filter=lfs diff=lfs merge=lfs -text
356
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_1545167900966233290+3cc9a3cb/model.neff filter=lfs diff=lfs merge=lfs -text
357
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_15eb37e3fb556fd3decb+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
358
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_1649fc77b87fff02e370+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
359
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_16885846486752037550+3cc9a3cb/model.neff filter=lfs diff=lfs merge=lfs -text
360
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_17147817242878056900+3cc9a3cb/model.neff filter=lfs diff=lfs merge=lfs -text
361
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_17215669526000154054+3cc9a3cb/model.neff filter=lfs diff=lfs merge=lfs -text
362
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_177e7d4bfa5feb0f544b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
363
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_17de9f35d8b452d30d64+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
364
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_185d416c669d1d2d5d27+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
365
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_18ea862943d46e9502fe/model.neuron filter=lfs diff=lfs merge=lfs -text
366
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_194792ca42c74e6c357a+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
367
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_19afba81b3e0cb8f888e+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
368
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_1af7783366403652ad51+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
369
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_2053b1e0fa543cbd84dd+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
370
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_20f9276a7330dc649731+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
371
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_215d5f9f03ff7f796870+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
372
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_226930e6ac9b200c488a+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
373
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_22cf23062ec53b3fd95d+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
374
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_23c20ba5fd4672262b4f+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
375
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_24ff9ac2787ce9a1d276+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
376
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_25c42f91982701aa03cc+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
377
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_264f2b3d508ee0e65f5f+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
378
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_272d7dabaeb48e6d7210+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
379
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_2a10c038669fa2d343e1+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
380
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_2a213e0741594f10219d+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
381
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_2ae37ee2c7255cce8028+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
382
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_2b6914194b931d7496fc+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
383
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_2b761d9812991a0ed26f+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
384
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_2d763f4e10d60c4030bd+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
385
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_2db40b5b23a523e16b87+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
386
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_2e229618015e416964c5+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
387
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_2e2fefac0330c92cf935+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
388
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_2ef52130792b59d66c66+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
389
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_2f59caeb45329072d7b0+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
390
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_2fa982b13af96f760a14+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
391
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_3043604d5429432a4ae1+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
392
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_3061bfbd0d476b938719+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
393
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_30a395d0dd4232fed110+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
394
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_320f2622d4d0c9fdd0f1+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
395
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_38847bfec695e36ba44b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
396
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_396f866b25e26394047c+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
397
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_399d134629eb34a00d43+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
398
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_3a070e0dd0e8091b4ebd+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
399
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_3aa61c04bd8d9ca8722a+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
400
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_3d7f2d2bef4f6fdd2c74+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
401
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_3d8b2e723048c421570a+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
402
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_3da832fdaa3d62981800+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
403
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_3e5f6b34247d2b457ec5+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
404
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_3f0110aa8aef5f42c4bc+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
405
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_3f45d3caa2f3862b756d+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
406
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_40b73cdf6e14999378ef+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
407
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_40c932379328aee8b619+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
408
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_40f92bf9469aae653e93+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
409
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_41bd479e657e66f398e1+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
410
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_4229851826662473223+3cc9a3cb/model.neff filter=lfs diff=lfs merge=lfs -text
411
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_45ee942ce275eb459e1d+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
412
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_4610853575678388020+3cc9a3cb/model.neff filter=lfs diff=lfs merge=lfs -text
413
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_4678455820992380476+3cc9a3cb/model.neff filter=lfs diff=lfs merge=lfs -text
414
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_4723142e6e2041791232+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
415
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_478acb1b0aae0b56f401+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
416
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_48437cc74469a8ccaec8+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
417
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_48bfe9ceb9631fdca2d4+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
418
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_48c6444adcac1b53fda6+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
419
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_498320a96768c9eab266+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
420
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_499a6f3c535d93bcd8c9/text_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
421
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_499a6f3c535d93bcd8c9/text_encoder_2/model.neuron filter=lfs diff=lfs merge=lfs -text
422
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_499a6f3c535d93bcd8c9/unet/model.neuron filter=lfs diff=lfs merge=lfs -text
423
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_499a6f3c535d93bcd8c9/vae_decoder/model.neuron filter=lfs diff=lfs merge=lfs -text
424
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_499a6f3c535d93bcd8c9/vae_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
425
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_4cccf279275b39cca49c+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
426
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_4dea33d55b39527f7fc4+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
427
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_4e6317a710ff2fcf60cd+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
428
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_4ed4eafdd8b57b2a58ba+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
429
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_4f54951d7f1d19705651+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
430
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_506e29dd5cc46918936d+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
431
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_51c384f52e045bea30bd+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
432
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_52108672c0a9ba2de711+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
433
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_531613db5d175a66b951+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
434
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_5371a46c9607cc9aeba7+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
435
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_53d5c248a142eced45e6+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
436
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_54947fa9c379486e4f5b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
437
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_56bfdcf572244e4771f8+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
438
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_59ddd6977b34718ba8b1+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
439
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_5a9cc28c7336fd9ea3ab+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
440
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_5b36e3b39e7f0fc4612b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
441
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_5b45c246024901dcd434+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
442
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_5de5ea910315ba9c0def+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
443
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_5e35dbcf8ed725f26ec0+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
444
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_626306e0d30994029849+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
445
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_6302731493553409576+3cc9a3cb/model.neff filter=lfs diff=lfs merge=lfs -text
446
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_66e776ad8f8401a7fa1f+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
447
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_6819e64fb96e87ffece0+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
448
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_682689dd0dabdac18fbd+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
449
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_686d3373c8eb03797f41+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
450
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_6886ba3bbca51509cd6c+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
451
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_6c9037784aa88d983971+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
452
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_6f479fa0ca80b88a5b46+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
453
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_7062a76356ca462bcc78+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
454
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_70e2b073f2bbeab1aac7/text_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
455
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_70e2b073f2bbeab1aac7/unet/model.neuron filter=lfs diff=lfs merge=lfs -text
456
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_70e2b073f2bbeab1aac7/vae_decoder/model.neuron filter=lfs diff=lfs merge=lfs -text
457
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_70e2b073f2bbeab1aac7/vae_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
458
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_71a034dd7c4a3afb59c6+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
459
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_7323a70130bb87727b84+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
460
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_7329da261de607372f14+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
461
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_73e7362a3d706e4803fc+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
462
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_7418d879b0b0dbe9d053+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
463
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_74a533aa34a2ab0fc7aa+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
464
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_75bc57af47ebdc36a75e+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
465
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_767b99765e404104c7bb+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
466
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_767d2f94879ca86f93cb+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
467
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_7ce64f3fbb3960eec319+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
468
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_7e2686740d19496c3c6e+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
469
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_7f54d4d3885e2a479ab4+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
470
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_7fce147fef3e3e4e1386+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
471
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_80ed60eb29ade17d45ff+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
472
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_827f458fd58fe6442166+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
473
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_838aaca91716f9f756f4+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
474
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_83cb40c0c38bacf5b8fd+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
475
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_843d9f5ba8d442721b5e/model.neuron filter=lfs diff=lfs merge=lfs -text
476
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_850d5a3d0694f6f65d91+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
477
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_857e6d9a3a28c7fe8c0c+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
478
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_867e29f5ed30739f8708+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
479
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_878607c2d60970a9fc41+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
480
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_88098bf708a0bb5dbef8/text_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
481
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_88098bf708a0bb5dbef8/unet/model.neuron filter=lfs diff=lfs merge=lfs -text
482
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_88098bf708a0bb5dbef8/vae_decoder/model.neuron filter=lfs diff=lfs merge=lfs -text
483
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_88098bf708a0bb5dbef8/vae_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
484
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_892a0bb27ce39228be75+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
485
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_897fc8ee0fee573d4294+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
486
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_899a6dab3f2005ac207a+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
487
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_8a11747c1cf643ac58ca+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
488
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_8a17d7b25a214e612693+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
489
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_8a3cac67f02909d50819+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
490
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_8b38e41cf0a3c0152b87+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
491
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_8b728c26e7384d3aa1fa+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
492
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_8bb2b91314df12f54a63+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
493
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_8d1befe3d92435e31fdf+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
494
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_8ea2de719456a80196f9+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
495
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_90381dde99583f54a60b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
496
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_905ba6adc472b9f33cc6+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
497
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_913f4e1e2b4632438fe9+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
498
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_922ce3591b342750be6b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
499
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_9294048697624734381+3cc9a3cb/model.neff filter=lfs diff=lfs merge=lfs -text
500
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_934e6cd7e79b2fea88ab+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
501
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_9402963399160769132+3cc9a3cb/model.neff filter=lfs diff=lfs merge=lfs -text
502
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_949152ff315aeb616749+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
503
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_94932be571ca32c0d6c9+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
504
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_94df5568a6ee6b8f721b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
505
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_94fd7c020f22384f6d27+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
506
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_95dae70c69a000bb4eb6+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
507
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_95f239593a81a9baed26+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
508
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_96624f9b555d2ee1123d+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
509
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_967d83f73b8fe30e4d14+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
510
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_9a2a6520ed76b8de514b+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
511
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_9ad3b35220e91074c83c+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
512
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_9b4b8c3000ad341d2ddc+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
513
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_9d4538dd1cfb96628706+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
514
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_9dc5393f49942ebdc1af+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
515
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_9dc8af288609e8374339+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
516
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_9e3cdf817e494ba86fd4+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
517
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_9f4e39af4e0b35889393+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
518
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_9f5a907ba7db9a6a1e51+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
519
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_a08a6102a96eea257396+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
520
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_a297084c31f9c1ff1d5d+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
521
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_a307a86e0d389d7d8d9e+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
522
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_a31729227d8ec6b4f136+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
523
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_a361a81ac38223fcadef+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
524
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_a6569d9c9d00e3812134+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
525
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_a71e46095542b60896d4/text_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
526
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_a71e46095542b60896d4/unet/model.neuron filter=lfs diff=lfs merge=lfs -text
527
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_a71e46095542b60896d4/vae_decoder/model.neuron filter=lfs diff=lfs merge=lfs -text
528
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_a71e46095542b60896d4/vae_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
529
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_a7513d3ab2f1b0306086+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
530
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_a923e74850212465e6ab+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
531
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_a95326667095c4d75db0+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
532
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_a9bcc20735f75ac1f46e+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
533
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_ab8029e291c37141655e+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
534
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_abb19eb6336ab05b7e19+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
535
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_ad7506c3fc27d75f26a6+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
536
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_aeecac19cd61e7fb1951+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
537
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_b2821092d30cdeb76d6a+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
538
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_b2aaa317ae645625c9d1+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
539
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_b3ac85ee34a5d9e564ba+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
540
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_b4a7a4856dcbd900d6ed+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
541
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_b9e80b41756cf2d7ed45+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
542
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_bdda47cac156ea8ffbfc+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
543
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_be5892d16c7e2fb6594c+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
544
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_bf70065b1150aaeca3bc+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
545
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_bf816c8623b1fd6b9185+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
546
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_c064389074584e41f54f+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
547
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_c0740fe961da69526f92+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
548
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_c1ff08289224c1071721+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
549
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_c24af7fcf05443daf3b7+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
550
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_c49c0c3715f68c22b32f+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
551
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_cb16b651ea9d180d5cfd+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
552
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_cc8cf237a3eedbaaeb9b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
553
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_cea91d96c873419351af+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
554
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_ceb1532721b05ed70905+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
555
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_d07d2aeb772cbb9c0ccd+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
556
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_d3e5e4b762fd9115b68b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
557
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_d6a9bb00aa7189af3ffb/text_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
558
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_d6a9bb00aa7189af3ffb/text_encoder_2/model.neuron filter=lfs diff=lfs merge=lfs -text
559
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_d6a9bb00aa7189af3ffb/unet/model.neuron filter=lfs diff=lfs merge=lfs -text
560
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_d6a9bb00aa7189af3ffb/vae_decoder/model.neuron filter=lfs diff=lfs merge=lfs -text
561
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_d6a9bb00aa7189af3ffb/vae_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
562
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_da84b070ab21ed1df54f+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
563
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_dcc643bd43691cdb2cd2+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
564
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_dd4a13f50d0dab4871b8+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
565
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_de5fab2e2ecc60551050+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
566
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_de8368a717cfd6dfec57+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
567
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_df48af4bf01af7f3857e+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
568
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_e0765cf6df2204e3664e+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
569
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_e10fffbed4e6a5dd8ea5+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
570
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_e40769e5a35d3e5dacff+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
571
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_e50c30ad9376ce18d039+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
572
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_e8eb6214f5c387ad6b43+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
573
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_e99dd53d1f163d6bb260+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
574
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_e9bcfc17d832317203bd+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
575
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_eaebd6f723294af70148+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
576
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_eca56defe775a7981c4b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
577
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_ef26a928b1a6003d9b67+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
578
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_f4c26fe7dbc0d9182bc8+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
579
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_f5b6904839bc118bcd54+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
580
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_f65495564d03f1c90e3a+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
581
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_f6b2b8267d631f2f0fad+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
582
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_f7a493b3a1ae24ab8ebb+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
583
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_f7bd275f6b204e1d6808+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
584
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_f914d922823d8ceace3c+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
585
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_f91cd98a64f373af274d+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
586
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_fa8e5e93b134884838d5+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
587
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_fadc62f786e9e6e1bf75+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
588
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_fce469267b2ad1b5d80e+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
589
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_fd6170cedb4fe53c8433+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
590
+ neuronxcc-2.16.372.0+4a9b2326/MODULE_feed1320e32bafc2a577+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
591
+ neuronxcc-2.17.194.0+d312836f/MODULE_1de63cf3d5b3c6c9921e/text_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
592
+ neuronxcc-2.17.194.0+d312836f/MODULE_1de63cf3d5b3c6c9921e/unet/model.neuron filter=lfs diff=lfs merge=lfs -text
593
+ neuronxcc-2.17.194.0+d312836f/MODULE_1de63cf3d5b3c6c9921e/vae_decoder/model.neuron filter=lfs diff=lfs merge=lfs -text
594
+ neuronxcc-2.17.194.0+d312836f/MODULE_1de63cf3d5b3c6c9921e/vae_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
595
+ neuronxcc-2.17.194.0+d312836f/MODULE_20223e04b1bfd1a5fe7e+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
596
+ neuronxcc-2.17.194.0+d312836f/MODULE_2e229618015e416964c5+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
597
+ neuronxcc-2.17.194.0+d312836f/MODULE_4f54951d7f1d19705651+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
598
+ neuronxcc-2.17.194.0+d312836f/MODULE_b9cbc3c9d2c2ba603243+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/gpt2/gpt2/780a42b2f8414c10d661.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"activation_function": "gelu_new", "architectures": ["GPT2LMHeadModel"], "attn_pdrop": 0.1, "bos_token_id": 50256, "embd_pdrop": 0.1, "eos_token_id": 50256, "initializer_range": 0.02, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 1024, "n_embd": 768, "n_head": 12, "n_inner": null, "n_layer": 12, "n_positions": 1024, "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "gpt2", "checkpoint_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 1024, "task": "text-generation"}, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "task_specific_params": {"text-generation": {"do_sample": true, "max_length": 50}}, "use_cache": true, "vocab_size": 50257}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/granite/ibm-granite/granite-3.1-2b-instruct/0211503669c9e68fbbde.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["GraniteForCausalLM"], "attention_bias": false, "attention_dropout": 0.1, "attention_multiplier": 0.015625, "bos_token_id": 0, "embedding_multiplier": 12.0, "eos_token_id": 0, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 8192, "logits_scaling": 8.0, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "granite", "neuron": {"auto_cast_type": "bf16", "batch_size": 2, "checkpoint_id": "ibm-granite/granite-3.1-2b-instruct", "checkpoint_revision": "374ef54e020a3ce208c65e96d6213922a87d8952", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 40, "num_key_value_heads": 8, "pad_token_id": 0, "residual_multiplier": 0.22, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 5000000.0, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 49155}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/granite/ibm-granite/granite-3.1-2b-instruct/7cf1a953f534c2649a7d.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["GraniteForCausalLM"], "attention_bias": false, "attention_dropout": 0.1, "attention_multiplier": 0.015625, "bos_token_id": 0, "embedding_multiplier": 12.0, "eos_token_id": 0, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 8192, "logits_scaling": 8.0, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "granite", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "ibm-granite/granite-3.1-2b-instruct", "checkpoint_revision": "07eae41278b4cd918dbe32da5d77c27e1379f751", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 40, "num_key_value_heads": 8, "pad_token_id": 0, "residual_multiplier": 0.22, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 5000000.0, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 49155}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/NousResearch/Hermes-2-Theta-Llama-3-8B/7bae97d51948b959db6b.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128003, "head_dim": 128, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "NousResearch/Hermes-2-Theta-Llama-3-8B", "checkpoint_revision": "57a73110702e7b05ba3f39fef36297454c680725", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/meta-llama/Llama-3.1-8B-Instruct/25395edbfe4cb40ed4e9.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 128, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "meta-llama/Llama-3.1-8B-Instruct", "checkpoint_revision": "0e9e39f249a16976918f6564b8830bc894c89659", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/meta-llama/Llama-3.2-1B-Instruct/7a87208395f3db5ba7be.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 64, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 8192, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "meta-llama/Llama-3.2-1B-Instruct", "checkpoint_revision": "9213176726f574b556790deb65791e0c5aa438b6", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 16, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 32.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/meta-llama/Llama-3.2-1B-Instruct/8cd73c74d69aa6440798.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 64, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 8192, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 2, "checkpoint_id": "meta-llama/Llama-3.2-1B-Instruct", "checkpoint_revision": "9213176726f574b556790deb65791e0c5aa438b6", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 16, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 32.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/meta-llama/Llama-3.2-1B-Instruct/ebbf84c3779811da8d2c.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 64, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 8192, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "meta-llama/Llama-3.2-1B-Instruct", "checkpoint_revision": "9213176726f574b556790deb65791e0c5aa438b6", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 16, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 32.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/meta-llama/Meta-Llama-3.1-8B-Instruct/3b6b0065e24ca9d2eeb0.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 128, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "meta-llama/Meta-Llama-3.1-8B-Instruct", "checkpoint_revision": "0e9e39f249a16976918f6564b8830bc894c89659", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/princeton-nlp/Sheared-LLaMA-1.3B/8f3a6f200be91cc32a57.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 128, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 5504, "max_position_embeddings": 4096, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "princeton-nlp/Sheared-LLaMA-1.3B", "checkpoint_revision": "a4b76938edbf571ea7d7d9904861cbdca08809b4", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 16, "num_hidden_layers": 24, "num_key_value_heads": 16, "pad_token_id": 0, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float32", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/mistral/optimum/mistral-1.1b-testing/6668357e5d6bb2f40fe7.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 64, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 5632, "max_position_embeddings": 32768, "model_type": "mistral", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "optimum/mistral-1.1b-testing", "checkpoint_revision": "ce03bc8d47dbd2c173ff65f3a8de1325ba724195", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 22, "num_key_value_heads": 4, "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-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/mixtral/dacorvo/Mixtral-tiny/f574775a377747e5836b.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["MixtralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 32, "hidden_act": "silu", "hidden_size": 1024, "initializer_range": 0.02, "intermediate_size": 3584, "max_position_embeddings": 1024, "model_type": "mixtral", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "dacorvo/Mixtral-tiny", "checkpoint_revision": "c557ba205ddff6ea911f4719e0d543d6c08356b6", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 1024, "task": "text-generation"}, "num_attention_heads": 32, "num_experts_per_tok": 2, "num_hidden_layers": 2, "num_key_value_heads": 8, "num_local_experts": 8, "output_router_logits": false, "rms_norm_eps": 1e-05, "rope_theta": 10000.0, "router_aux_loss_coef": 0.001, "router_jitter_noise": 0.0, "sliding_window": 4096, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-3-medium-4k-instruct/ae83aa0fea9a9d0315d5.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {"AutoConfig": "microsoft/Phi-3-medium-4k-instruct--configuration_phi3.Phi3Config", "AutoModelForCausalLM": "microsoft/Phi-3-medium-4k-instruct--modeling_phi3.Phi3ForCausalLM"}, "bos_token_id": 1, "embd_pdrop": 0.0, "eos_token_id": 32000, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 4096, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "microsoft/Phi-3-medium-4k-instruct", "checkpoint_revision": "ae004ae82eb6eddc32906dfacb1d6dfea8f91996", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 4096, "pad_token_id": 32000, "partial_rotary_factor": 1.0, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "sliding_window": 2047, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32064}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-3-mini-4k-instruct/13bb5f11971eace7eeed.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {"AutoConfig": "microsoft/Phi-3-mini-4k-instruct--configuration_phi3.Phi3Config", "AutoModelForCausalLM": "microsoft/Phi-3-mini-4k-instruct--modeling_phi3.Phi3ForCausalLM"}, "bos_token_id": 1, "embd_pdrop": 0.0, "eos_token_id": 32000, "hidden_act": "silu", "hidden_size": 3072, "initializer_range": 0.02, "intermediate_size": 8192, "max_position_embeddings": 4096, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "microsoft/Phi-3-mini-4k-instruct", "checkpoint_revision": "0a67737cc96d2554230f90338b163bc6380a2a85", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "original_max_position_embeddings": 4096, "pad_token_id": 32000, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "sliding_window": 2047, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32064}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-3-mini-4k-instruct/29904a0c0edb4679ee76.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {"AutoConfig": "microsoft/Phi-3-mini-4k-instruct--configuration_phi3.Phi3Config", "AutoModelForCausalLM": "microsoft/Phi-3-mini-4k-instruct--modeling_phi3.Phi3ForCausalLM"}, "bos_token_id": 1, "embd_pdrop": 0.0, "eos_token_id": 32000, "hidden_act": "silu", "hidden_size": 3072, "initializer_range": 0.02, "intermediate_size": 8192, "max_position_embeddings": 4096, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "microsoft/Phi-3-mini-4k-instruct", "checkpoint_revision": "0a67737cc96d2554230f90338b163bc6380a2a85", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "original_max_position_embeddings": 4096, "pad_token_id": 32000, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "sliding_window": 2047, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32064}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-3-mini-4k-instruct/9a957125ff99f484a66e.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {"AutoConfig": "microsoft/Phi-3-mini-4k-instruct--configuration_phi3.Phi3Config", "AutoModelForCausalLM": "microsoft/Phi-3-mini-4k-instruct--modeling_phi3.Phi3ForCausalLM"}, "bos_token_id": 1, "embd_pdrop": 0.0, "eos_token_id": 32000, "hidden_act": "silu", "hidden_size": 3072, "initializer_range": 0.02, "intermediate_size": 8192, "max_position_embeddings": 4096, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "microsoft/Phi-3-mini-4k-instruct", "checkpoint_revision": "0a67737cc96d2554230f90338b163bc6380a2a85", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "original_max_position_embeddings": 4096, "pad_token_id": 32000, "partial_rotary_factor": 1.0, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "sliding_window": 2047, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32064}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-3-mini-4k-instruct/d7892042a1da5eaa0020.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {"AutoConfig": "microsoft/Phi-3-mini-4k-instruct--configuration_phi3.Phi3Config", "AutoModelForCausalLM": "microsoft/Phi-3-mini-4k-instruct--modeling_phi3.Phi3ForCausalLM"}, "bos_token_id": 1, "embd_pdrop": 0.0, "eos_token_id": 32000, "hidden_act": "silu", "hidden_size": 3072, "initializer_range": 0.02, "intermediate_size": 8192, "max_position_embeddings": 4096, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 2, "checkpoint_id": "microsoft/Phi-3-mini-4k-instruct", "checkpoint_revision": "0a67737cc96d2554230f90338b163bc6380a2a85", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "original_max_position_embeddings": 4096, "pad_token_id": 32000, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "sliding_window": 2047, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32064}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-4-mini-instruct/6ff6f9ae27117fd24bd7.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {"AutoConfig": "microsoft/Phi-4-mini-instruct--configuration_phi3.Phi3Config", "AutoModelForCausalLM": "microsoft/Phi-4-mini-instruct--modeling_phi3.Phi3ForCausalLM", "AutoTokenizer": "microsoft/Phi-4-mini-instruct--Xenova/gpt-4o"}, "bos_token_id": 199999, "embd_pdrop": 0.0, "eos_token_id": 199999, "full_attn_mod": 1, "hidden_act": "silu", "hidden_size": 3072, "initializer_range": 0.02, "intermediate_size": 8192, "interpolate_factor": 1, "lm_head_bias": false, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "microsoft/Phi-4-mini-instruct", "checkpoint_revision": "4b00ec8714b0cb224e4fb33380cbf0919f177f3e", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 24, "num_hidden_layers": 32, "num_key_value_heads": 8, "original_max_position_embeddings": 4096, "pad_token_id": 199999, "partial_rotary_factor": 0.75, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": {"long_factor": [1, 1.118320672, 1.250641126, 1.398617824, 1.564103225, 1.74916897, 1.956131817, 2.187582649, 2.446418898, 2.735880826, 3.059592084, 3.421605075, 3.826451687, 4.279200023, 4.785517845, 5.351743533, 5.984965424, 6.693110555, 7.485043894, 8.370679318, 9.36110372, 10.4687158, 11.70738129, 13.09260651, 14.64173252, 16.37415215, 18.31155283, 20.47818807, 22.90118105, 25.61086418, 28.64115884, 32.03, 32.1, 32.13, 32.23, 32.6, 32.61, 32.64, 32.66, 32.7, 32.71, 32.93, 32.97, 33.28, 33.49, 33.5, 44.16, 47.77], "short_factor": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "type": "longrope"}, "rope_theta": 10000.0, "sliding_window": 262144, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 200064}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/42644c76f78cfd9108b4.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100265, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "microsoft/phi-4", "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 16384, "pad_token_id": 100349, "partial_rotary_factor": 1.0, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/452310f61d209c411ec5.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100265, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "microsoft/phi-4", "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 16384, "pad_token_id": 100349, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/6faff6d0347b9b3197ed.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100265, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 2, "checkpoint_id": "microsoft/phi-4", "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 16384, "pad_token_id": 100349, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/aefe5e89e89e5ba40a59.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100265, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 24, "checkpoint_id": "microsoft/phi-4", "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 10, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 16384, "pad_token_id": 100349, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/b05887d807c6f53298d2.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100265, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 16, "checkpoint_id": "microsoft/phi-4", "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 16384, "pad_token_id": 100349, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/c488c1faf1a22992a2ea.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100265, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 16, "checkpoint_id": "microsoft/phi-4", "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 16384, "pad_token_id": 100349, "partial_rotary_factor": 1.0, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/e21879dbd09a21fb7a2e.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100265, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 8, "checkpoint_id": "microsoft/phi-4", "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 10, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 16384, "pad_token_id": 100349, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/f6232f240b5b98678a51.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100265, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 16, "checkpoint_id": "microsoft/phi-4", "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 12, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 16384, "pad_token_id": 100349, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/yujiepan/phi-4-tiny-random/33891c5b973ec2bfb226.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {}, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100257, "hidden_act": "silu", "hidden_size": 16, "initializer_range": 0.02, "intermediate_size": 32, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "yujiepan/phi-4-tiny-random", "checkpoint_revision": "18a9a1168dc97ac6d128f811925670c275610f5a", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 2, "num_hidden_layers": 2, "num_key_value_heads": 1, "original_max_position_embeddings": 16384, "pad_token_id": 100257, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/yujiepan/phi-4-tiny-random/48d7f1c1e412ea84b60e.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {}, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100257, "hidden_act": "silu", "hidden_size": 16, "initializer_range": 0.02, "intermediate_size": 32, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "yujiepan/phi-4-tiny-random", "checkpoint_revision": "18a9a1168dc97ac6d128f811925670c275610f5a", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 2, "num_hidden_layers": 2, "num_key_value_heads": 1, "original_max_position_embeddings": 16384, "pad_token_id": 100257, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/yujiepan/phi-4-tiny-random/a37d77e276a4a0f5c4e8.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {}, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100257, "hidden_act": "silu", "hidden_size": 16, "initializer_range": 0.02, "intermediate_size": 32, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "yujiepan/phi-4-tiny-random", "checkpoint_revision": "18a9a1168dc97ac6d128f811925670c275610f5a", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 2, "num_hidden_layers": 2, "num_key_value_heads": 1, "original_max_position_embeddings": 16384, "pad_token_id": 100257, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/qwen2/Qwen/Qwen2.5-0.5B/08b82ea5dab18178ad7f.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": "2.16.372.0+4a9b2326", "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_scaling": null, "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-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/qwen2/Qwen/Qwen2.5-Math-1.5B/a5b11f2dd182bd813171.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": 1536, "initializer_range": 0.02, "intermediate_size": 8960, "max_position_embeddings": 4096, "max_window_layers": 21, "model_type": "qwen2", "neuron": {"auto_cast_type": "bf16", "batch_size": 2, "checkpoint_id": "Qwen/Qwen2.5-Math-1.5B", "checkpoint_revision": "4a83ca6e4526a4f2da3aa259ec36c259f66b2ab2", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 12, "num_hidden_layers": 28, "num_key_value_heads": 2, "rms_norm_eps": 1e-06, "rope_scaling": null, "rope_theta": 10000, "sliding_window": null, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "use_mrope": false, "use_sliding_window": false, "vocab_size": 151936}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/058e3ead348125e7808b.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"apply_residual_connection_post_layernorm": false, "architectures": ["BloomForCausalLM"], "attention_dropout": 0.1, "bos_token_id": 1, "dtype": "float32", "eos_token_id": 2, "gradient_checkpointing": false, "hidden_dropout": 0.1, "hidden_size": 32, "id2label": {"0": "LABEL_0", "1": "LABEL_1", "2": "LABEL_2"}, "initializer_range": 0.02, "is_decoder": true, "label2id": {"LABEL_0": 0, "LABEL_1": 1, "LABEL_2": 2}, "layer_norm_epsilon": 1e-05, "model_type": "bloom", "n_head": 4, "n_layer": 5, "n_positions": 512, "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "hf-internal-testing/tiny-random-BloomForCausalLM", "checkpoint_revision": "92b07e9b7b4f986fa7c54e2ac3b9201b4ba5212e", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "pad_token_id": 3, "pretraining_tp": 1, "seq_length": 7, "slow_but_exact": true, "torch_dtype": "float32", "type_vocab_size": 16, "use_cache": true, "vocab_size": 1024}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/6bf9ef24cedd2d181630.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"apply_residual_connection_post_layernorm": false, "architectures": ["BloomForCausalLM"], "attention_dropout": 0.1, "bos_token_id": 1, "dtype": "float32", "eos_token_id": 2, "gradient_checkpointing": false, "hidden_dropout": 0.1, "hidden_size": 32, "id2label": {"0": "LABEL_0", "1": "LABEL_1", "2": "LABEL_2"}, "initializer_range": 0.02, "is_decoder": true, "label2id": {"LABEL_0": 0, "LABEL_1": 1, "LABEL_2": 2}, "layer_norm_epsilon": 1e-05, "model_type": "bloom", "n_head": 4, "n_layer": 5, "n_positions": 512, "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "hf-internal-testing/tiny-random-BloomForCausalLM", "checkpoint_revision": "92b07e9b7b4f986fa7c54e2ac3b9201b4ba5212e", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "pad_token_id": 3, "pretraining_tp": 1, "seq_length": 7, "slow_but_exact": true, "torch_dtype": "float32", "type_vocab_size": 16, "use_cache": true, "vocab_size": 1024}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/7a4c62323b876a6ad082.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"apply_residual_connection_post_layernorm": false, "architectures": ["BloomForCausalLM"], "attention_dropout": 0.1, "bos_token_id": 1, "dtype": "float32", "eos_token_id": 2, "gradient_checkpointing": false, "hidden_dropout": 0.1, "hidden_size": 32, "id2label": {"0": "LABEL_0", "1": "LABEL_1", "2": "LABEL_2"}, "initializer_range": 0.02, "is_decoder": true, "label2id": {"LABEL_0": 0, "LABEL_1": 1, "LABEL_2": 2}, "layer_norm_epsilon": 1e-05, "model_type": "bloom", "n_head": 4, "n_layer": 5, "n_positions": 512, "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "hf-internal-testing/tiny-random-BloomForCausalLM", "checkpoint_revision": "92b07e9b7b4f986fa7c54e2ac3b9201b4ba5212e", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "pad_token_id": 3, "pretraining_tp": 1, "seq_length": 7, "slow_but_exact": true, "torch_dtype": "float32", "type_vocab_size": 16, "use_cache": true, "vocab_size": 1024}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/gpt2/gpt2/780a42b2f8414c10d661.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"activation_function": "gelu_new", "architectures": ["GPT2LMHeadModel"], "attn_pdrop": 0.1, "bos_token_id": 50256, "embd_pdrop": 0.1, "eos_token_id": 50256, "initializer_range": 0.02, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 1024, "n_embd": 768, "n_head": 12, "n_inner": null, "n_layer": 12, "n_positions": 1024, "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "gpt2", "checkpoint_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 1024, "task": "text-generation"}, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "task_specific_params": {"text-generation": {"do_sample": true, "max_length": 50}}, "use_cache": true, "vocab_size": 50257}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/gpt2/hf-internal-testing/tiny-random-gpt2/3033d959070fc853146f.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"activation_function": "gelu_new", "attention_probs_dropout_prob": 0.1, "attn_pdrop": 0.1, "bos_token_id": 98, "embd_pdrop": 0.1, "eos_token_id": 98, "gradient_checkpointing": false, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "initializer_range": 0.02, "intermediate_size": 37, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 512, "n_embd": 32, "n_head": 4, "n_inner": null, "n_layer": 5, "n_positions": 512, "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "hf-internal-testing/tiny-random-gpt2", "checkpoint_revision": "71034c5d8bde858ff824298bdedc65515b97d2b9", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "pad_token_id": 98, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "type_vocab_size": 16, "use_cache": true, "vocab_size": 1000}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/gpt2/hf-internal-testing/tiny-random-gpt2/46dc72bec418e48dc0b2.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"activation_function": "gelu_new", "attention_probs_dropout_prob": 0.1, "attn_pdrop": 0.1, "bos_token_id": 98, "embd_pdrop": 0.1, "eos_token_id": 98, "gradient_checkpointing": false, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "initializer_range": 0.02, "intermediate_size": 37, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 512, "n_embd": 32, "n_head": 4, "n_inner": null, "n_layer": 5, "n_positions": 512, "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "hf-internal-testing/tiny-random-gpt2", "checkpoint_revision": "71034c5d8bde858ff824298bdedc65515b97d2b9", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "pad_token_id": 98, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "type_vocab_size": 16, "use_cache": true, "vocab_size": 1000}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/gpt2/hf-internal-testing/tiny-random-gpt2/fdfe8a25cf9f41629cb9.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"activation_function": "gelu_new", "attention_probs_dropout_prob": 0.1, "attn_pdrop": 0.1, "bos_token_id": 98, "embd_pdrop": 0.1, "eos_token_id": 98, "gradient_checkpointing": false, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "initializer_range": 0.02, "intermediate_size": 37, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 512, "n_embd": 32, "n_head": 4, "n_inner": null, "n_layer": 5, "n_positions": 512, "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "hf-internal-testing/tiny-random-gpt2", "checkpoint_revision": "71034c5d8bde858ff824298bdedc65515b97d2b9", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "pad_token_id": 98, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "type_vocab_size": 16, "use_cache": true, "vocab_size": 1000}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/granite/ibm-granite/granite-3.1-2b-instruct/bf71032506a251068193.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["GraniteForCausalLM"], "attention_bias": false, "attention_dropout": 0.1, "attention_multiplier": 0.015625, "bos_token_id": 0, "embedding_multiplier": 12.0, "eos_token_id": 0, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 8192, "logits_scaling": 8.0, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "granite", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "ibm-granite/granite-3.1-2b-instruct", "checkpoint_revision": "374ef54e020a3ce208c65e96d6213922a87d8952", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 40, "num_key_value_heads": 8, "pad_token_id": 0, "residual_multiplier": 0.22, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 5000000.0, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 49155}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/llama/Maykeye/TinyLLama-v0/f71b7fdfb0dfed1fac13.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 4, "hidden_act": "silu", "hidden_size": 64, "initializer_range": 0.02, "intermediate_size": 256, "max_position_embeddings": 2048, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 2, "checkpoint_id": "Maykeye/TinyLLama-v0", "checkpoint_revision": "298338802ab94432b917bcce11382aa151aee50f", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 1024, "task": "text-generation"}, "num_attention_heads": 16, "num_hidden_layers": 8, "num_key_value_heads": 16, "pad_token_id": 0, "pretraining_tp": 1, "rms_norm_eps": 1e-06, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32000}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/llama/NousResearch/Hermes-2-Theta-Llama-3-8B/7bae97d51948b959db6b.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128003, "head_dim": 128, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 8192, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "NousResearch/Hermes-2-Theta-Llama-3-8B", "checkpoint_revision": "57a73110702e7b05ba3f39fef36297454c680725", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/llama/llamafactory/tiny-random-Llama-3/2e8cb42c92e334dfbaa9.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 4, "hidden_act": "silu", "hidden_size": 16, "initializer_range": 0.02, "intermediate_size": 64, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "llamafactory/tiny-random-Llama-3", "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 2, "num_key_value_heads": 4, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 128256}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/llama/llamafactory/tiny-random-Llama-3/5433fffb45752137c8de.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 4, "hidden_act": "silu", "hidden_size": 16, "initializer_range": 0.02, "intermediate_size": 64, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "llamafactory/tiny-random-Llama-3", "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 2, "num_key_value_heads": 4, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 128256}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/llama/llamafactory/tiny-random-Llama-3/8c48d1da1325809763e8.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 4, "hidden_act": "silu", "hidden_size": 16, "initializer_range": 0.02, "intermediate_size": 64, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "llamafactory/tiny-random-Llama-3", "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 2, "num_key_value_heads": 4, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 128256}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/llama/unsloth/Llama-3.2-1B-Instruct/49d61e18927b113552d1.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": 128009, "head_dim": 64, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 8192, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "unsloth/Llama-3.2-1B-Instruct", "checkpoint_revision": "9b58d4a36161a1e49ecf0a69d20b2736fef8e438", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 16, "num_key_value_heads": 8, "pad_token_id": 128004, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 32.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "unsloth_fixed": true, "use_cache": true, "vocab_size": 128256}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/3afe2ef31d28f1e456be.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "attention_probs_dropout_prob": 0.1, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 8, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "hidden_size": 32, "initializer_range": 0.02, "intermediate_size": 37, "is_decoder": true, "max_position_embeddings": 512, "model_type": "mistral", "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "dacorvo/tiny-random-MistralForCausalLM", "checkpoint_revision": "81d453e3c8985649e9ee3d4c9378461029d1c73a", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 2, "num_key_value_heads": 2, "pad_token_id": 0, "rms_norm_eps": 1e-06, "rope_theta": 10000.0, "sliding_window": 4096, "tie_word_embeddings": false, "torch_dtype": "float32", "type_vocab_size": 16, "use_cache": true, "vocab_size": 32000}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/aa9d07318c2ee53a79c9.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "attention_probs_dropout_prob": 0.1, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 8, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "hidden_size": 32, "initializer_range": 0.02, "intermediate_size": 37, "is_decoder": true, "max_position_embeddings": 512, "model_type": "mistral", "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "dacorvo/tiny-random-MistralForCausalLM", "checkpoint_revision": "81d453e3c8985649e9ee3d4c9378461029d1c73a", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 2, "num_key_value_heads": 2, "pad_token_id": 0, "rms_norm_eps": 1e-06, "rope_theta": 10000.0, "sliding_window": 4096, "tie_word_embeddings": false, "torch_dtype": "float32", "type_vocab_size": 16, "use_cache": true, "vocab_size": 32000}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/mistral/dacorvo/tiny-random-MistralForCausalLM/ef87caaa517e593a73ba.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "attention_probs_dropout_prob": 0.1, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 8, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "hidden_size": 32, "initializer_range": 0.02, "intermediate_size": 37, "is_decoder": true, "max_position_embeddings": 512, "model_type": "mistral", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "dacorvo/tiny-random-MistralForCausalLM", "checkpoint_revision": "81d453e3c8985649e9ee3d4c9378461029d1c73a", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 2, "num_key_value_heads": 2, "pad_token_id": 0, "rms_norm_eps": 1e-06, "rope_theta": 10000.0, "sliding_window": 4096, "tie_word_embeddings": false, "torch_dtype": "float32", "type_vocab_size": 16, "use_cache": true, "vocab_size": 32000}
neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/mistral/optimum/mistral-1.1b-testing/6668357e5d6bb2f40fe7.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"architectures": ["MistralForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 64, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 5632, "max_position_embeddings": 32768, "model_type": "mistral", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "optimum/mistral-1.1b-testing", "checkpoint_revision": "ce03bc8d47dbd2c173ff65f3a8de1325ba724195", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 22, "num_key_value_heads": 4, "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}