2026-05-11 UTC
# 
superkuh LLM are deterministic too up to the point of the sampler selecting from logits. The actual neural network part will always return the same set of logits for the same input. It's the normal, algorithmic (not neural network style) bolt-ons to chose *not* the most probable logit that introduces indeterminism. But not really because with the same sampler settings you always get the same. Now, corporate hosted services are bad because you have no contr