Meta and UvA Introduce Context-weighted Discrete Flow Matching
Meta and the University of Amsterdam proposed Context-weighted Discrete Flow Matching to improve discrete generative models. By refining traditional DFM, the method preserves parallel generation capabilities while reducing perplexity by 63% on OpenWebText.
2026-07-26 ~ 2026-07-26 · 2 related posts
- Context-weighted flow matching cuts perplexity 63% on OpenWebText — burny_tech · 2026-07-26
- Meta and UvA improve discrete flow matching with easier-token prioritization — burkov · 2026-07-26