currently LLMs from majorvproviders are not deterministic with temp=0, there are startups focusing on this issue (among others) https://thinkingmachines.ai/blog/defeating-nondeterminism-in...