Defeating Nondeterminism in LLM Inference

LLM이 $t=0.0$인 경우에도 대체로 비결정론적인 이유를 상세히 설명.

Reproducibility is a bedrock of scientific progress. However, it’s remarkably difficult to get reproducible results out of large language models.

thinkingmachines.ai/blog/defeating-nondeterminism-in-llm-inference/