LLMs output probability distribution of the next token. And that automatically makes them non-deterministic. You can make their output deterministic by greedy s

LLMs output probability distribution of the next token. And that automatically makes them non-deterministic. You can make their output deterministic by greedy sampling, fixing the seed of a pseudorandom generator, or by computing exponentially growing probability distribution of all possible continuations, but it doesn't change the fact that LLMs produce probability distributions you need to sample somehow to get a definite result.