Soila Kavulya

icon indicating an email [email protected]

Results 3 issues of Soila Kavulya

Errors loading DDPO model finetuned with LORA after saving locally

I am experiencing problems loading the DDPO model saved locally using `trainer.save_pretrained` after finetuning with LoRA. There are errors about missing keys. ``` Traceback (most recent call last): File "ddpo_eval.py",...

Add support for contrastive search

11 comment

# What does this PR do? Adds support for contrastive search for static and dynamic inputs, and low memory configs. Also adds support for `GPT2DoubleHeadsModel`. Fixes the tests below: ```...

run-test
synapse1.17

Add support for optimum-habana deepseek v3/r1 fp8 quantization

## Type of Change # What does this PR do? Support FP8 static quantization for optimum-habana deepseek v3/r1 models using Intel Neural Compressor (INC) This feature needs changes in: -...