Teaching Diffusion to Speculate Left-to-Right
作者: Lexington Whalen, Yuki Ito, Ryo Sakamoto
分类: cs.CL, cs.LG
发布日期: 2026-06-10
备注: 13 pages, technical report
💡 一句话要点
提出三种干预措施以优化扩散语言模型的左到右推理
🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)
关键词: 扩散语言模型 投机解码 自回归模型 损失函数设计 自然语言处理 文本生成 推理效率
📋 核心要点
- 现有的自回归解码方法在推理时存在显著的顺序限制，导致推理效率低下。
- 论文提出通过标记位置加权、首错误聚焦损失和链损失等三种干预措施来优化扩散语言模型的生成过程。
- 实验结果表明，这三种干预措施在多个基准上显著提高了接受草稿长度，提升幅度达到21-76%。
📝 摘要（中文）
大型语言模型（LLMs）在多种任务中表现出色，但其自回归解码过程导致推理成本高昂。投机解码通过使用轻量级草稿模型并行生成多个未来标记来解决这一瓶颈。最近的研究表明，扩散语言模型适合这一场景，因为它们可以并行生成整个草稿标记块，从而减轻自回归草稿的顺序约束。本文通过实证分析三种训练时间干预措施，缩小了草稿生成与验证之间的差距，分别是标记位置加权、针对每个块中打破接受前缀位置的首错误聚焦损失，以及替代期望接受长度的链损失项。这三种干预措施在四个目标模型和六个基准上提高了接受草稿长度21-76%。
🔬 方法详解
问题定义：本文旨在解决扩散语言模型在生成草稿标记时与自回归模型验证之间的顺序不一致问题。现有方法在推理过程中存在效率低下和生成质量不一致的挑战。
核心思路：论文提出三种干预措施，旨在通过调整训练过程来缩小草稿生成与验证之间的差距，从而提高生成效率和质量。
技术框架：整体架构包括草稿生成模块和验证模块。草稿生成模块使用扩散模型并行生成标记块，验证模块则使用自回归模型对生成的标记进行逐个验证。
关键创新：最重要的技术创新在于提出了三种干预措施，这些措施分别针对标记位置、首错误和接受前缀的联合优化，能够有效提升生成质量。
关键设计：在损失函数设计上，采用了首错误聚焦损失和链损失，确保模型在训练时能够更好地适应验证阶段的需求，同时保持了原有推理管道的完整性。实验中未增加额外的前向传递，确保了推理效率。
🖼️ 关键图片
📊 实验亮点
实验结果显示，三种干预措施在四个目标模型和六个基准上显著提高了接受草稿长度，提升幅度达21-76%，且未增加额外的前向传递，保持了推理管道的高效性。
🎯 应用场景
该研究的潜在应用领域包括自然语言处理中的文本生成、对话系统和代码生成等。通过优化推理过程，能够显著提高模型的响应速度和生成质量，具有广泛的实际价值和未来影响。
📄 摘要（原文）
Large language models (LLMs) achieve remarkable performance across a wide range of tasks, but their autoregressive decoding process incurs substantial inference costs due to inherently sequential token generation. Speculative decoding addresses this bottleneck by employing a lightweight draft model to propose multiple future tokens that are subsequently verified in parallel by a larger target model. Recent work has demonstrated that diffusion language models are well suited for this setting, as they can generate entire blocks of draft tokens in parallel and thereby alleviate the sequential constraints of autoregressive drafting. A subtlety of this regime is that block-diffusion drafters generate tokens bidirectionally within a block, whereas verification is performed by an autoregressive target model that evaluates tokens in a strictly left-to-right manner, leaving a gap between the symmetric training-time objective and the asymmetric verification-time reward. In this work, we offer an empirical analysis of three training-time interventions that narrow this gap: token positional weighting, a first-error focal loss that targets the position that breaks the accepted prefix within each block, and a chain loss term that substitutes a differentiable surrogate for the expected accepted length. The three interventions act along orthogonal axes (position, block-conditional first error, joint prefix) and compose additively; they are likewise orthogonal to test-time alignment mechanisms such as multi-draft self-selection, with which they can in principle be combined. Across four target models and six reasoning, code, and dialogue benchmarks, the three interventions raise accepted draft length by 21-76% per benchmark over a position-uniform baseline, without adding additional forward passes and without changing the inference pipeline or the rejection-sampling exactness contract.