Meta and UvA improve discrete flow matching with easier-token prioritization
burkov · x · 2026-07-26
Researchers from @AIatMeta and @UvAAmsterdam propose a way to make diffusion-style discrete generation more practical by prioritizing easier, better-supported token positions during generation and training. The method preserves any-order generation, can be applied to a pretrained model with no extra training and almost no overhead, and reduces generative perplexity by up to 63% on OpenWebText while also improving validity and novelty in molecule generation.
Related event: Meta and UvA Introduce Context-weighted Discrete Flow Matching(2 posts)→
More from Models
- OpenAI ships GPT-6.1 Terra; Brockman says its computer use now beats his own speed — gdb · 2026-10-27
- Kimi K3.1 Pro regresses on Debate Benchmark while DeepSeek 1017 gains at lower cost — teortaxesTex · 2026-10-27
- 4 Hours of LLM Startup Research Produced a Convincing but Nonsense Product Thesis — annetgriffin · 2026-10-27
- Dev argues Claude burns far more tokens than GPT models for the same coding work — haider1 · 2026-10-27
- MiniMax matches Claude Opus 4.5 in non-benchmaxxed evals, users report — teortaxesTex · 2026-10-27
- Real-world math: GPT-6.1 Terra's half price is offset by 2-3x token usage, costing more than Sol — lordpuddingcup · 2026-10-27