`包裹,长文用`
`/`
`划分小节。 3. 列表内容用`
- `,引用用`
`,代码用`
`,重要句子可用``。
4. 保留原文所有信息,不得增删改内容。
5. 不要输出``/``/``等文档级标签。
正文内容:
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.
分析正文结构:
这是一段关于掩码扩散语言模型和ReMaRL方法的学术/技术新闻。
可以划分为几个小节来体现``或`
`,或者直接用`
`。由于文本不长,可以分为几个自然段,或者加一个小标题。
比如:
`,代码用``,重要句子可用``。 4. 保留原文所有信息,不得增删改内容。 5. 不要输出``/``/``等文档级标签。 正文内容: 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. 分析正文结构: 这是一段关于掩码扩散语言模型和ReMaRL方法的学术/技术新闻。 可以划分为几个小节来体现``或`
`,或者直接用`
`。由于文本不长,可以分为几个自然段,或者加一个小标题。 比如:
背景与现有方法
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.
ReMaRL方法与实验结果
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.
检查是否增删改了内容: 原文: 1. Masked diffusion language models generate tokens in parallel, and the order in which tokens are unmasked affects final-answer accuracy. 2. Several methods learn unmasking policies from task rewards, but these policies cannot remask earlier predictions. 3. Remasking lets the model correct those predictions using newly generated context. 4. Heuristic rules guide both unmasking and remasking, while supervised remaskers learn to detect token errors. 5. Neither approach directly optimizes these decisions for final-answer correctness. 6. To address this gap, we introduce ReMaRL (Remasking with Reinforcement Learning), a lightweight policy that learns which tokens to unmask, retain, or remask. 7. We train this roughly 0.3M-parameter policy with Group Relative Policy Optimization (GRPO) from final-answer rewards while keeping the diffusion model frozen. 8. 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. 排版后的HTML:研究背景与现有方法
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.
ReMaRL方法与实验成果
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.
等等,规则说“不得增删改内容”。我添加了小标题“研究背景与现有方法”和“ReMaRL方法与实验成果”,这算不算“增”内容? 规则4:“保留原文所有信息,不得增删改内容”。添加标题可能会被认为是增加了原文没有的文字。 为了严格遵守“不得增删改内容”,最好不使用自创的标题文字,或者只使用HTML结构标签而不添加原文没有的可见文本。 但是规则2说“长文用/
划分小节”。如果不添加标题,怎么划分小节? 可以将原文分成两个`
`,或者用`
`包裹原文中的某句话作为小节标题?原文没有明显的小节标题。 或者,我可以直接用`
`包裹整个段落,或者分成几个`
`,不强行加`
`。因为原文只有一段,或者可以按句群分成两段。 原文: 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. 分成两个自然段:
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