arXiv cs.CL· Lexington Whalen, Yuki Ito, Ryo Sakamoto·· 12 小时前AI 评分31
让扩散模型学会从左到右推测：三项训练干预缩小投机解码的验证鸿沟
Teaching Diffusion to Speculate Left-to-Right
AI 导读
针对扩散草稿模型块内双向生成与自回归目标模型严格从左到右验证之间的不对称，研究提出 token 位置加权、首个错误焦点损失和链式损失三项训练干预，在四个目标模型、六个推理/代码/对话基准上把接受草稿长度较位置均匀基线提升 21-76%，且不增加额外前向计算、不改变推理流程与拒绝采样精确性契约。
正文 · 原文
Abstract: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.
| Comments: | 13 pages, technical report |
| Subjects: | Computation and Language (cs.CL); Machine Learning (cs.LG) |
| Cite as: | arXiv:2606.11552 [cs.CL] |
| (or arXiv:2606.11552v3 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2606.11552 arXiv-issued DOI via DataCite |
Submission history
From: Lexington Whalen [view email]
[v1]
Wed, 10 Jun 2026 01:21:56 UTC (1,311 KB)
[v2]
Mon, 22 Jun 2026 23:07:36 UTC (1,311 KB)
[v3]
Wed, 4 Nov 2026 18:51:59 UTC (2,243 KB)
来源：arXiv cs.CL · arxiv.org