Ask HN: GPT non-deterministic, feature or bug?
Seems like GPT non-determinism is mostly from sparse MoE.
Imagine if you use a DeepMind approach of tree traversal for the most promising paths, it can generate deterministic LLM responses.
Imagine if you use a DeepMind approach of tree traversal for the most promising paths, it can generate deterministic LLM responses.
No comments yet.