`包裹,长文用`
`/`
`划分小节(这里文本不长,可以视情况加,或者不加,但为了体现专业性,可以加一个合适的`
`作为标题或小节,或者就按段落来。原文只有一段,可以拆分成几个逻辑段落,或者保持一段,但为了符合“长文用h2/h3”,这里虽然不算特别长,但可以加个h2作为小节标题,比如“研究背景与V-lens”或者干脆不加,只用p。仔细看,原文就是一段完整的摘要。为了严格遵守“保留原文所有信息,不得增删改内容”,我不应该随便添加原文没有的标题文字。所以不加h2/h3,或者只把原文按句子群分成几个p。 等等,规则说“长文用
/
划分小节”,如果原文没有明确的小节,强行加标题会违反“不得增删改内容”。所以我只使用`
`,或者把原文合理分段。 原文: "Visualizing intermediate representations could help us inspect the reasoning steps underlying a model output. Unified multimodal models offer this opportunity by combining understanding and generation in one architecture. Their image generators can access representations formed while processing text. We introduce the V-lens, which uses this connection to visualize a chosen token's hidden state without training a decoder. Across four diffusion-based and two purely autoregressive unified models, the resulting images reveal key intermediate concepts that connect a question to its answer even when the concepts are never named explicitly. Controlled tests show how this content changes with context, token position, and depth. We then introduce lens tuning, which supervises selected hidden states with pictures through the V-lens. Across BAGEL, SenseNova-U1, and HunyuanImage-3.0, the rate of correct visualizations of pronoun referents rises from 2-11% to 86-94%. Surprisingly, such visual supervision also improves text answers without supervising the answer tokens directly. Code is available at https://github.com/multimodal-ai-lab/V-lens." 为了排版美观,可以将其分为几个段落,或者就一个段落。既然规则说“每个自然段落用
包裹”,原文是一整段,那我就用一个`
`包裹。或者按逻辑拆分成几个`
`,但“不得增删改内容”意味着不能改变文字,换行算不算改内容?通常拆分自然段是可以的,但最安全的是保持为一个`
`,或者按句子逻辑拆成几个`
`。 让我们按逻辑拆分成几个`
`: 段落1:Visualizing intermediate representations ... processing text. 段落2:We introduce the V-lens ... named explicitly. 段落3:Controlled tests show ... through the V-lens. 段落4:Across BAGEL ... directly. 段落5:Code is available ... 或者就整体作为一个`
`。原文就是一段,保持一段最稳妥,不增删任何内容(包括换行符)。 不过,为了体现“排版工程师”的专业感,将长段落按语义拆分为几个`
`是合理的排版操作,且没有增删改文字内容。 拆分:
Visualizing intermediate representations could help us inspect the reasoning steps underlying a model output. Unified multimodal models offer this opportunity by combining understanding and generation in one architecture. Their image generators can access representations formed while processing text.
We introduce the V-lens, which uses this connection to visualize a chosen token's hidden state without training a decoder. Across four diffusion-based and two purely autoregressive unified models, the resulting images reveal key intermediate concepts that connect a question to its answer even when the concepts are never named explicitly.
Controlled tests show how this content changes with context, token position, and depth. We then introduce lens tuning, which supervises selected hidden states with pictures through the V-lens.
Across BAGEL, SenseNova-U1, and HunyuanImage-3.0, the rate of correct visualizations of pronoun referents rises from 2-11% to 86-94%. Surprisingly, such visual supervision also improves text answers without supervising the answer tokens directly.
Code is available at https://github.com/multimodal-ai-lab/V-lens.
等等,规则说“不得增删改内容”,加``标签算不算改内容?把URL变成链接是语义化HTML的常规操作,但严格来说改变了文本结构。如果不用``,就纯文本:Code is available at https://github.com/multimodal-ai-lab/V-lens.
或者用``包裹URL?规则说“代码用
”,URL不算代码。所以直接纯文本即可。
关于重要句子可用``:
"We introduce the V-lens" -> V-lens 可以用 `V-lens`
"lens tuning" -> `lens tuning`
检查规则:
1. 只输出HTML片段本身,不要Markdown围栏、不要解释;
2. 每个自然段落用包裹,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。
原文没有列表、引用、代码(除了URL,但URL不是代码块)。
所以只需要``和``。
段落划分:
Visualizing intermediate representations could help us inspect the reasoning steps underlying a model output. Unified multimodal models offer this opportunity by combining understanding and generation in one architecture. Their image generators can access representations formed while processing text.
We introduce the V-lens, which uses this connection to visualize a chosen token's hidden state without training a decoder. Across four diffusion-based and two purely autoregressive unified models, the resulting images reveal key intermediate concepts that connect a question to its answer even when the concepts are never named explicitly.
Controlled tests show how this content changes with context, token position, and depth. We then introduce lens tuning, which supervises selected hidden states with pictures through the V-lens.
Across BAGEL, SenseNova-U1, and HunyuanImage-3.0, the rate of correct visualizations of pronoun referents rises from 2-11% to 86-94%. Surprisingly, such visual supervision also improves text answers without supervising the answer tokens directly.
Code is available at https://github.com/multimodal-ai-lab/V-lens.
这样既符合排版习惯,又没有增删改任何文字内容(除了添加HTML标签)。
再核对一遍文字:
Visualizing intermediate representations could help us inspect the reasoning steps underlying a model output. Unified multimodal models offer this opportunity by combining understanding and generation in one architecture. Their image generators can access representations formed while processing text.
We introduce the V-lens, which uses this connection to visualize a chosen token's hidden state without training a decoder. Across four diffusion-based and two purely autoregressive unified models, the resulting images reveal key intermediate concepts that connect a question to its answer even when the concepts are never named explicitly.
Controlled tests show how this content changes with context, token position, and depth. We then introduce lens tuning, which supervises selected hidden states with pictures through the V-lens.
Across BAGEL, SenseNova-U1, and HunyuanImage-3.0, the rate of correct visualizations of pronoun referents rises from 2-11% to 86-94%. Surprisingly, such visual supervision also improves text answers without supervising the answer tokens directly.
Code is available at https://github.com