包裹，看看正文结构，先通读一遍内容。 首先，第一句到“However, aggressively quantized models often perform poorly on reasoning tasks.”是第一个自然段吗？不对，应该看整体结构。然后“We observe larger activation fluctuations during the first decoding steps of reasoning. Calibration that omits these early states may miss their activation fluctuations.”这两句是不是一段？接下来“To address these reasoning-specific quantization challenges, we propose \textbf{CodeQuant2}...”这里是提出CodeQuant2，里面有几个方法，需要看有没有需要列表的？然后后面描述CodeQuant2的几个措施，还有最终效果。 首先，先处理每个自然段落。原文中的结构： 第一段：Large reasoning models (LRMs) improve task performance by generating longer intermediate reasoning traces, but this substantially increases inference costs and makes low-precision quantization increasingly important. However, aggressively quantized models often perform poorly on reasoning tasks. 第二段：We observe larger activation fluctuations during the first decoding steps of reasoning. Calibration that omits these early states may miss their activation fluctuations. 第三段：To address these reasoning-specific quantization challenges, we propose \textbf{CodeQuant2}, a clustering-based LUT quantization framework for low-precision large reasoning models. 第四段：CodeQuant2 mitigates activation outliers with rotation and scaling, uses clustering to represent non-uniform weight distributions, and adapts codebook sizes to the quantization difficulty of each layer. 第五段：Guided by our observation of reasoning-time activation behavior, CodeQuant2 further improves calibration sample selection to better match actual decoding statistics. 第六段：Combined with our customized LUT CUDA kernel, CodeQuant2 delivers substantial accuracy improvements on reasoning benchmarks while achieving up to $2.85\times$ speedup over BF16. 不过，规则里说列表内容用
- ，看看有没有需要列表的地方？第四段里的三个措施：mitigates activation outliers...、uses clustering to represent...、adapts codebook sizes...这三个是并列的，是不是可以作为列表？对，因为是并列的方法，所以可以用
- 来呈现，这样更清晰，而且符合规则。 然后，\textbf{CodeQuant2}里的\textbf要转换成吗？因为规则里说重要句子可用，这里的CodeQuant2是重要的模型名称，所以应该把\textbf{CodeQuant2}改成CodeQuant2。 然后，检查有没有需要
/
的小节？原文是比较连贯的，可能不需要分小节，因为没有明显的小节标题，所以直接用
和必要的列表、。 现在整理： 首先第一段：
Large reasoning models (LRMs) improve task performance by generating longer intermediate reasoning traces, but this substantially increases inference costs and makes low-precision quantization increasingly important. However, aggressively quantized models often perform poorly on reasoning tasks.
第二段：We observe larger activation fluctuations during the first decoding steps of reasoning. Calibration that omits these early states may miss their activation fluctuations.
第三段：To address these reasoning-specific quantization challenges, we propose CodeQuant2, a clustering-based LUT quantization framework for low-precision large reasoning models.
第四段原来的三个并列措施，改成列表：CodeQuant2 employs multiple strategies to address quantization challenges:
- mitigates activation outliers with rotation and scaling
- uses clustering to represent non-uniform weight distributions
- adapts codebook sizes to the quantization difficulty of each layer
Guided by our observation of reasoning-time activation behavior, CodeQuant2 further improves calibration sample selection to better match actual decoding statistics.
第六段：Combined with our customized LUT CUDA kernel, CodeQuant2 delivers substantial accuracy improvements on reasoning benchmarks while achieving up to $2.85\times$ speedup over BF16.
等等，但是原文第四段是“CodeQuant2 mitigates activation outliers with rotation and scaling, uses clustering to represent non-uniform weight distributions, and adapts codebook sizes to the quantization difficulty of each layer.”如果改成列表的话，是不是要保持一致？不过规则说保留原文所有信息，不得增删改内容，所以不能加“employs multiple strategies to address quantization challenges:”这样的句子，因为原文没有，所以不能加。那怎么办？原文的三个动作是并列的，用逗号连接，如果用列表的话，是不是需要把句子改成列表形式，但不能增删内容？ 哦，规则4说保留原文所有信息，不得增删改内容，所以不能添加额外的话。那原来的这句话是一个并列句，用逗号分隔，所以可能不能改成列表，因为没有额外的引导语，不然就增了内容。那只能用包裹整个句子，因为如果强行拆分成列表，要么需要加引导语（增内容），要么把原句拆成三个
- ，但每个
- 里的内容和原句的结构不符，会改变内容吗？ 比如原句是“CodeQuant2 mitigates activation outliers with rotation and scaling, uses clustering to represent non-uniform weight distributions, and adapts codebook sizes to the quantization difficulty of each layer.”如果拆成三个
- ，每个
- 对应一个动作，主语还是CodeQuant2，那每个
- 应该是：
- CodeQuant2 mitigates activation outliers with rotation and scaling
- CodeQuant2 uses clustering to represent non-uniform weight distributions
- CodeQuant2 adapts codebook sizes to the quantization difficulty of each layer
这样的话，原句的三个动作被拆分成三个句子，补充了主语？不，原句的三个动作都是CodeQuant2发出的，所以拆分成每个- 都带主语的话，是不是改变了原句的结构？原句是一个句子，用and连接最后一个动作，所以拆分成列表的话，相当于把一个句子拆成三个独立的分句，但是内容是完整的，没有增删，只是结构变了？ 不过规则里说列表内容用
- ，如果原句是并列的动作，属于列表内容吗？规则里说“列表内容”，可能指原文中就是列表形式的，或者明确并列的内容。如果原文是一个句子中的并列谓语，可能不需要拆成列表，否则会改变原文的结构，违反规则4。 那再仔细看规则4：保留原文所有信息，不得增删改内容。所以不能改变句子结构吗？比如原句是一个句子，不能拆分成多个句子或列表，否则就是改了内容结构？ 那回到原文，有没有都是非列表内容，所以全部用
包裹，只处理和必要的标签。 那第三段里的\textbf{CodeQuant2}要改成CodeQuant2，因为\textbf是LaTeX的加粗，对应HTML的，这一步是对的，因为原文的CodeQuant2需要加粗处理，不然信息没保留。 然后检查有没有其他需要处理的？比如$2.85\times$是数学公式，规则里说代码用
，但数学公式不属于代码，所以不需要处理，保留原样，因为规则里没有说处理数学公式，只说不增删改内容，所以$2.85\times$就保留，不用改成HTML的公式标签，因为规则里没要求。 现在再整理每个段落： 第一段：Large reasoning models (LRMs) improve task performance by generating longer intermediate reasoning traces, but this substantially increases inference costs and makes low-precision quantization increasingly important. However, aggressively quantized models often perform poorly on reasoning tasks.
第二段：We observe larger activation fluctuations during the first decoding steps of reasoning. Calibration that omits these early states may miss their activation fluctuations.
第三段：To address these reasoning-specific quantization challenges, we propose CodeQuant2, a clustering-based LUT quantization framework for low-precision large reasoning models.
第四段：CodeQuant2 mitigates activation outliers with rotation and scaling, uses clustering to represent non-uniform weight distributions, and adapts codebook sizes to the quantization difficulty of each layer.
第五段：Guided by our observation of reasoning
- 来呈现，这样更清晰，而且符合规则。 然后，\textbf{CodeQuant2}里的\textbf要转换成吗？因为规则里说重要句子可用，这里的CodeQuant2是重要的模型名称，所以应该把\textbf{CodeQuant2}改成CodeQuant2。 然后，检查有没有需要