fix RLHF llama rewarding modeling backward issue#612

regisss merged 2 commits into

mainhuggingface/optimum-habana:mainfrom

reward_llamahuggingface/optimum-habana:reward_llamaCopy head branch name to clipboard