Thinking Machines Lab analyzes causes of LLM inference nondeterminism and open-sources batch-invariant kernels
Original title:Defeating Nondeterminism in LLM Inference
AI summary
Thinking Machines Lab published a blog post arguing that the real cause of LLM inference nondeterminism is not concurrency plus floating point, but the lack of batch invariance in kernels: batch size changes driven by server load alter the output of an individual request.
Quick Facts
- Published
- Updated
- Source
- Thinking Machines Lab:官方博客
- Category
- Papers & Research
- AI score
- 68
作者指出 LLM 推理不确定性的真因是 batch size 变化而非并发浮点问题,并给出可复现内核方案。
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Thinking Machines Lab analyzes causes of LLM inference nondeterminism and open-sources batch-invariant kernels. September 10, 2025. https://new.nat6.net/en/news/cmq75sy1q0009sl5w22bhu5g3/