there's a huggingface embedding leaderboard, may not have been focused on llama for performance reasons - https://huggingface.co/spaces/mteb/leaderboard might be able to tune llama(2) for recommendation systems using something like this: https://github.com/SAI990323/TALLRec