← 研踪 PaperTrailarXiv:2611.22576 · v1
ReMaRL: Learning to Unmask and Remask in Frozen Diffusion Language Models
首次提交 2026-11-17 19:01
版本更新 2026-11-17 19:01
雷达观测 2026-11-18 11:33
摘要
Masked diffusion language models generate tokens in parallel, and the order in which tokens are unmasked affects final-answer accuracy. Several methods learn unmasking policies from task rewards, but these policies cannot remask earlier predictions. Remasking lets the model correct those predictions using newly generated context. Heuristic rules guide both unmasking and remasking, while supervised remaskers learn to detect token errors. Neither approach directly optimizes these decisions for final-answer correctness. To address this gap, we introduce ReMaRL (Remasking with Reinforcement Learning), a lightweight policy that learns which tokens to unmask, retain, or remask. We train this roughly 0.3M-parameter policy with Group Relative Policy Optimization (GRPO) from final-answer rewards while keeping the diffusion model frozen. With LLaDA-8B-Instruct, ReMaRL achieves the highest scores among the compared unmasking and remasking policies within the tested inference budgets of 4-32 model calls on math and 8-16 calls on code.
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v1 · 2026-11-17 19:01 · 最新观测版本
ReMaRL: Learning to Unmask and Remask in Frozen Diffusion Language Models
Masked diffusion language models generate tokens in parallel, and the order in which tokens are unmasked affects final-answer accuracy. Several methods learn unmasking policies from task rewards, but these policies cannot remask earlier predictions. Remasking lets the model correct those predictions using newly generated context. Heuristic rules guide both unmasking and remasking, while supervised remaskers learn to detect token errors. Neither approach directly optimizes these decisions for final-answer correctness. To address this gap, we introduce ReMaRL (Remasking with Reinforcement Learning), a lightweight policy that learns which tokens to unmask, retain, or remask. We train this roughly 0.3M-parameter policy with Group Relative Policy Optimization (GRPO) from final-answer rewards while keeping the diffusion model frozen. With LLaDA-8B-Instruct, ReMaRL achieves the highest scores among the compared unmasking and remasking policies within the tested inference budgets of 4-32 model calls on math and 8-16 calls on code.
查看此版本解读 →查看此版本原文 ↗