Flide: open-source inference engine with exact state forking claims up to 9.5x speedup over vLLM
_akhaliq · x · 2026-11-11
Flide is a new open-source inference engine for Jev-like decision models, built around exact state forking, which lets multiple queries share computation while preserving exact decision semantics.
- Up to 9.5x faster inference than vLLM and SGLang in the authors' tests
- Supports 30+ open-source decision models on Decision Index 0.4.1, with more coming
- Install with one line: pip install flide
- Code, docs, and the arXiv paper (2610.42148) are all public
More from Infra
- Pretraining insider notes: data repetition and ~1000 TPP overtraining are standard practice — SonglinYang4 · 2026-11-11
- Stripe acquires OpenRouter to build the economic infrastructure for AI — jeff_weinstein · 2026-11-11
- NVDA's two Black Mondays: -17% on DeepSeek shock vs only -4.3% this time — cfregly · 2026-11-11
- Engineer turns NCCL Broadcast into a meme about Jensen Huang's apology — cfregly · 2026-11-11
- __popc counts set bits: Lori's 31 beats Alice's single high bit — cfregly · 2026-11-11
- Blackwell's 2-CTA MMA: one thread issues, two thread blocks compute — cfregly · 2026-11-11