`包裹，长文用`
`/`
`划分小节（这里内容不长，可能不需要小节，或者分几个小节）。 3. 列表用`
- `，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. 不要输出文档级标签。
原文：
Remasking allows diffusion language models to revisit generated text, but effective correction requires knowing when and what to revise. Our diagnosis on Sudoku reveals two gaps. Hidden states contain error information that token confidence does not reliably expose, and remasking an isolated error usually reproduces it when the surrounding context remains fixed. These findings motivate DiVR , a verifier that predicts whether a partially generated sequence will reach a correct answer. Trained on the final correctness of the model's own trajectories, DiVR decides when to regenerate a block and which alternative to retain, while the generator itself stays unchanged. On five reasoning benchmarks with LLaDA-8B-Base, DiVR raises the average accuracy of the fine-tuned generator from 35.3\% to 43.4\% at $1.9$-$3.5\times$ the cost of greedy decoding, above search baselines that use $3.8$-$7.4\times$. It also complements post-training such as supervised fine-tuning and reinforcement learning, and generalizes to other generators, including Dream.
分析原文结构，可以将其分为几个逻辑部分，并使用适当的HTML标签：
段落1：介绍背景和诊断发现。
"Remasking allows diffusion language models to revisit generated text, but effective correction requires knowing when and what to revise. Our diagnosis on Sudoku reveals two gaps." -> 用``包裹，可以把“knowing when and what to revise”加粗。
段落2：具体说明两个gaps。
"Hidden states contain error information that token confidence does not reliably expose, and remasking an isolated error usually reproduces it when the surrounding context remains fixed." -> 可以加粗两个gap的核心，或者作为列表？原文是一句话，直接用`
`，其中“Hidden states contain error information that token confidence does not reliably expose”和“remasking an isolated error usually reproduces it when the surrounding context remains fixed”可以用`
- `来表示两个gaps，但原文是“and”连接的一句话。为了严格遵守“不得增删改内容”，最好保持原句结构，或者把两个部分作为列表？如果改成列表，算改变内容结构吗？“保留原文所有信息，不得增删改内容”，通常指文字内容。如果保持句子完整，用`
`包裹并加粗关键点更稳妥。
段落3：引出DiVR。
"These findings motivate DiVR , a verifier that predicts whether a partially generated sequence will reach a correct answer." -> `
`，加粗DiVR。
段落4：DiVR的训练和机制。
"Trained on the final correctness of the model's own trajectories, DiVR decides when to regenerate a block and which alternative to retain, while the generator itself stays unchanged." -> `
`，加粗关键动作。
段落5：实验结果。
"On five reasoning benchmarks with LLaDA-8B-Base, DiVR raises the average accuracy of the fine-tuned generator from 35.3\% to 43.4\% at $1.9$-$3.5\times$ the cost of greedy decoding, above search baselines that use $3.8$-$7.4\times$." -> `
`，这里面的数学公式和百分比，用户要求代码用`
`，但这里不是代码，是数学公式，用``包裹数学公式可能不合适，或者保持原样，或者用``？规则说“代码用``”，这里不是代码，所以保持普通文本，或者把"35.3\%"等用``或``？最好保持纯文本，因为不是代码。
段落6：补充说明。
"It also complements post-training such as supervised fine-tuning and reinforcement learning, and generalizes to other generators, including Dream." -> ``，加粗“complements post-training”等。
为了体现“长文用`
`/`
`划分小节”，虽然这篇不算长文，但我可以添加适当的小节标题以符合规则的提示，或者不添加，因为不是长文。规则是“长文用`
`/`
`划分小节”，这里不是长文，所以可以不用。但如果加上小节，也是可以的。为了安全，不加小节标题，直接用`
`。
让我们仔细检查原文，看是否有可以拆分成列表的地方。
"Hidden states contain error information that token confidence does not reliably expose, and remasking an isolated error usually reproduces it when the surrounding context remains fixed."
这可以拆成：
- Hidden states contain error information that token confidence does not reliably expose
- remasking an isolated error usually reproduces it when the surrounding context remains fixed
但是原文是一句话用"and"连接。拆成列表会改变标点（去掉逗号和and），属于“增删改内容”。所以不拆成列表，保持原句，用``强调。
排版设计：
Remasking allows diffusion language models to revisit generated text, but effective correction requires knowing when and what to revise. Our diagnosis on Sudoku reveals two gaps.
Hidden states contain error information that token confidence does not reliably expose, and remasking an isolated error usually reproduces it when the surrounding context remains fixed.
These findings motivate DiVR, a verifier that predicts whether a partially generated sequence will reach a correct answer.
Trained on the final correctness of the model's own trajectories, DiVR decides when to regenerate a block and which alternative to retain, while the generator itself stays unchanged.
On five reasoning benchmarks with LLaDA-8B-Base, DiVR raises the average accuracy of the fine-tuned generator from 35.3\% to 43.4\% at $1.9$-$3.5\times$ the cost of greedy decoding, above search baselines that use $3.8$-$7.4\times$.
It also complements post-training such as supervised fine-tuning and reinforcement learning, and generalizes to other generators, including Dream.
检查是否有代码：没有。
检查是否有引用：没有。
检查是否有列表：没有（不适合强行拆分）。
确保没有Markdown围栏，没有解释，只输出HTML。
注意原文中的空格："DiVR , a verifier" -> 原文有空格"DiVR ,"，我应该保留吗？“不得增删改内容”，所以保留"DiVR ,"或者修正为"DiVR,"？通常排版可以修正明显的笔误，但规则说“不得增删改内容”，严格保留"DiVR ,"。
原文："DiVR , a verifier"
原文："35.3\% to 43.4\%" -> 保留反斜杠。
原文："$1.9$-$3.5\times$" -> 保留美元符号。
修改后的HTML：
Remasking allows diffusion language models to revisit generated text, but effective correction requires knowing when and what to revise. Our diagnosis on Sudoku reveals two gaps.
Hidden states contain error information that token confidence does not reliably expose, and remasking an isolated error usually reproduces it when the surrounding context remains fixed.
These findings motivate DiVR , a verifier that predicts whether a partially generated sequence will reach a correct answer.
Trained on the final correctness of the model's own trajectories, DiVR decides when to regenerate
`，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. 不要输出文档级标签。 原文： Remasking allows diffusion language models to revisit generated text, but effective correction requires knowing when and what to revise. Our diagnosis on Sudoku reveals two gaps. Hidden states contain error information that token confidence does not reliably expose, and remasking an isolated error usually reproduces it when the surrounding context remains fixed. These findings motivate DiVR , a verifier that predicts whether a partially generated sequence will reach a correct answer. Trained on the final correctness of the model's own trajectories, DiVR decides when to regenerate a block and which alternative to retain, while the generator itself stays unchanged. On five reasoning benchmarks with LLaDA-8B-Base, DiVR raises the average accuracy of the fine-tuned generator from 35.3\% to 43.4\% at $1.9$-$3.5\times$ the cost of greedy decoding, above search baselines that use $3.8$-$7.4\times$. It also complements post-training such as supervised fine-tuning and reinforcement learning, and generalizes to other generators, including Dream. 分析原文结构，可以将其分为几个逻辑部分，并使用适当的HTML标签： 段落1：介绍背景和诊断发现。 "Remasking allows diffusion language models to revisit generated text, but effective correction requires knowing when and what to revise. Our diagnosis on Sudoku reveals two gaps." -> 用``包裹，可以把“knowing when and what to revise”加粗。 段落2：具体说明两个gaps。 "Hidden states contain error information that token confidence does not reliably expose, and remasking an isolated error usually reproduces it when the surrounding context remains fixed." -> 可以加粗两个gap的核心，或者作为列表？原文是一句话，直接用`
`，其中“Hidden states contain error information that token confidence does not reliably expose”和“remasking an isolated error usually reproduces it when the surrounding context remains fixed”可以用`
- `来表示两个gaps，但原文是“and”连接的一句话。为了严格遵守“不得增删改内容”，最好保持原句结构，或者把两个部分作为列表？如果改成列表，算改变内容结构吗？“保留原文所有信息，不得增删改内容”，通常指文字内容。如果保持句子完整，用`
`包裹并加粗关键点更稳妥。 段落3：引出DiVR。 "These findings motivate DiVR , a verifier that predicts whether a partially generated sequence will reach a correct answer." -> `
`，加粗DiVR。 段落4：DiVR的训练和机制。 "Trained on the final correctness of the model's own trajectories, DiVR decides when to regenerate a block and which alternative to retain, while the generator itself stays unchanged." -> `
`，加粗关键动作。 段落5：实验结果。 "On five reasoning benchmarks with LLaDA-8B-Base, DiVR raises the average accuracy of the fine-tuned generator from 35.3\% to 43.4\% at $1.9$-$3.5\times$ the cost of greedy decoding, above search baselines that use $3.8$-$7.4\times$." -> `
`，这里面的数学公式和百分比，用户要求代码用`
`，但这里不是代码，是数学公式，用``包裹数学公式可能不合适，或者保持原样，或者用``？规则说“代码用``”，这里不是代码，所以保持普通文本，或者把"35.3\%"等用``或``？最好保持纯文本，因为不是代码。 段落6：补充说明。 "It also complements post-training such as supervised fine-tuning and reinforcement learning, and generalizes to other generators, including Dream." -> ``，加粗“complements post-training”等。 为了体现“长文用`
`/`
`划分小节”，虽然这篇不算长文，但我可以添加适当的小节标题以符合规则的提示，或者不添加，因为不是长文。规则是“长文用`
`/`
`划分小节”，这里不是长文，所以可以不用。但如果加上小节，也是可以的。为了安全，不加小节标题，直接用`
`。 让我们仔细检查原文，看是否有可以拆分成列表的地方。 "Hidden states contain error information that token confidence does not reliably expose, and remasking an isolated error usually reproduces it when the surrounding context remains fixed." 这可以拆成：
但是原文是一句话用"and"连接。拆成列表会改变标点（去掉逗号和and），属于“增删改内容”。所以不拆成列表，保持原句，用``强调。 排版设计：
- Hidden states contain error information that token confidence does not reliably expose
- remasking an isolated error usually reproduces it when the surrounding context remains fixed
Remasking allows diffusion language models to revisit generated text, but effective correction requires knowing when and what to revise. Our diagnosis on Sudoku reveals two gaps.
Hidden states contain error information that token confidence does not reliably expose, and remasking an isolated error usually reproduces it when the surrounding context remains fixed.
These findings motivate DiVR, a verifier that predicts whether a partially generated sequence will reach a correct answer.
Trained on the final correctness of the model's own trajectories, DiVR decides when to regenerate a block and which alternative to retain, while the generator itself stays unchanged.
On five reasoning benchmarks with LLaDA-8B-Base, DiVR raises the average accuracy of the fine-tuned generator from 35.3\% to 43.4\% at $1.9$-$3.5\times$ the cost of greedy decoding, above search baselines that use $3.8$-$7.4\times$.
It also complements post-training such as supervised fine-tuning and reinforcement learning, and generalizes to other generators, including Dream.
检查是否有代码：没有。 检查是否有引用：没有。 检查是否有列表：没有（不适合强行拆分）。 确保没有Markdown围栏，没有解释，只输出HTML。 注意原文中的空格："DiVR , a verifier" -> 原文有空格"DiVR ,"，我应该保留吗？“不得增删改内容”，所以保留"DiVR ,"或者修正为"DiVR,"？通常排版可以修正明显的笔误，但规则说“不得增删改内容”，严格保留"DiVR ,"。 原文："DiVR , a verifier" 原文："35.3\% to 43.4\%" -> 保留反斜杠。 原文："$1.9$-$3.5\times$" -> 保留美元符号。 修改后的HTML：Remasking allows diffusion language models to revisit generated text, but effective correction requires knowing when and what to revise. Our diagnosis on Sudoku reveals two gaps.
Hidden states contain error information that token confidence does not reliably expose, and remasking an isolated error usually reproduces it when the surrounding context remains fixed.
These findings motivate DiVR , a verifier that predicts whether a partially generated sequence will reach a correct answer.
Trained on the final correctness of the model's own trajectories, DiVR decides when to regenerate