Research Papers 论文研究 3h ago Updated 46m ago 更新于 46分钟前 44

Speculative Correction: Draft-then-Refine Decoding for Diffusion Language Models 推测校正:扩散语言模型的先草稿后精炼解码

Introduces a plug-and-play "draft-then-refine" decoding pattern for diffusion language models (DLMs) that first generates a complete draft, then refines it bidirectionally Flash-Flash configuration (same model drafting and refining) improves GSM8K-384 accuracy from 0.848 to 0.899 and MBPP-384 from 0.545 to 0.693, while running 1.20x faster than the block-autoregressive baseline Mini-Flash speculative correction configuration achieves MATH-384 performance of 0.294 versus 0.300 for Flash while run 介绍了一种即插即用的"先起草后精炼"解码模式,用于扩散语言模型(DLMs),该模式首先生成完整草稿,然后进行双向精炼。Flash-Flash配置(同一模型负责起草和精炼)将GSM8K-384的准确率从0.848提升至0.899,MBPP-384从0.545提升至0.693,同时比块自回归基线快1.20倍。Mini-Flash推测性修正配置在运行速度比Flash快2.17倍的情况下,在MATH-384上达到0.294的性能(对比Flash的0.300),证明了无需训练的加速路径。因果消融实验证实,完整草稿提供了有用的初始化,全局精炼有利于GSM8K,局部精炼在MBPP和MATH上获得收益。

62
Hot 热度
68
Quality 质量
58
Impact 影响力

Analysis 深度分析

TL;DR

  • Introduces a plug-and-play "draft-then-refine" decoding pattern for diffusion language models (DLMs) that first generates a complete draft, then refines it bidirectionally
  • Flash-Flash configuration (same model drafting and refining) improves GSM8K-384 accuracy from 0.848 to 0.899 and MBPP-384 from 0.545 to 0.693, while running 1.20x faster than the block-autoregressive baseline
  • Mini-Flash speculative correction configuration achieves MATH-384 performance of 0.294 versus 0.300 for Flash while running 2.17x faster, demonstrating a training-free speedup route
  • Causal ablations confirm that completed drafts provide useful initializations, with full global refinement benefiting GSM8K and local refinement capturing gains on MBPP and MATH

Why It Matters

This work demonstrates that bidirectional refinement is a genuinely useful decoding primitive for diffusion language models, not just a theoretical property. The draft-then-refine paradigm offers practitioners a simple, training-free inference-time optimization that can simultaneously improve accuracy and reduce latency, challenging the assumption that DLMs must be adapted to purely left-to-right generation.

Technical Details

  • Flash-Flash configuration: The same LLaDA2.1-Flash model serves as both drafter (block-autoregressive generation) and refiner (bidirectional diffusion refinement), testing whether a model can improve its own autoregressive output through global refinement
  • Mini-Flash configuration: Inspired by speculative decoding, LLaDA2.1-Mini drafts a full response and LLaDA2.1-Flash revises it as an editable initialization, enabling heterogeneous model cascades without additional training
  • Benchmark results: Flash-Flash achieves GSM8K-384: 0.899 (up from 0.848), MBPP-384: 0.693 (up from 0.545); Mini-Flash achieves MATH-384: 0.294 vs Flash's 0.300 at 2.17x speedup
  • Ablation findings: Refinement from fully masked spans performs poorly; completed drafts are essential; full global refinement provides clear GSM8K gains; local refinement captures much of the MBPP and MATH improvement
  • Latency-window-matched controls confirm gains persist after targeted tuning of block-autoregressive decoding, ruling out trivial speed-accuracy tradeoffs

Industry Insight

  • The draft-then-refine pattern offers an immediate, training-free optimization for any deployed diffusion language model, potentially improving both quality and throughput without retraining
  • Speculative correction between model sizes (Mini drafting, Flash refining) provides a practical recipe for building fast DLM inference pipelines that approach full-model quality at significantly reduced latency
  • Results support a Pareto-frontier interpretation rather than uniform quality matching, suggesting practitioners should carefully calibrate draft-refine configurations for their specific quality-latency requirements rather than assuming heterogeneous cascades will match the larger model

摘要

介绍了一种即插即用的"先起草后精炼"解码模式,用于扩散语言模型(DLMs),该模式首先生成完整草稿,然后进行双向精炼。Flash-Flash配置(同一模型负责起草和精炼)将GSM8K-384的准确率从0.848提升至0.899,MBPP-384从0.545提升至0.693,同时比块自回归基线快1.20倍。Mini-Flash推测性修正配置在运行速度比Flash快2.17倍的情况下,在MATH-384上达到0.294的性能(对比Flash的0.300),证明了无需训练的加速路径。因果消融实验证实,完整草稿提供了有用的初始化,全局精炼有利于GSM8K,局部精炼在MBPP和MATH上获得收益。

深度分析

一句话总结

  • 介绍了一种即插即用的"先起草后精炼"解码模式,用于扩散语言模型(DLMs),该模式首先生成完整草稿,然后进行双向精炼
  • Flash-Flash配置(同一模型负责起草和精炼)将GSM8K-384的准确率从0.848提升至0.899,MBPP-384从0.545提升至0.693,同时比块自回归基线快1.20倍
  • Mini-Flash推测性修正配置在运行速度比Flash快2.17倍的情况下,在MATH-384上达到0.294的性能(对比Flash的0.300),证明了无需训练的加速路径
  • 因果消融实验证实,完整草稿提供了有用的初始化,全局精炼有利于GSM8K,局部精炼在MBPP和MATH上获得收益

重要性

这项工作证明了双向精炼是扩散语言模型真正有用的解码原语,而不仅仅是理论特性。"先起草后精炼"范式为实践者提供了一种简单、无需训练的推理时优化方法,可以同时提高准确性和降低延迟,挑战了DLMs必须适应纯从左到右生成的假设。

技术细节

  • Flash-Flash配置:相同的LLaDA2.1-Flash模型同时充当起草者(块自回归生成)和精炼者(双向扩散精炼),测试模型是否可以通过全局精炼改进自身的自回归输出
  • Mini-Flash配置:受推测解码启发,LLaDA2.1-Mini起草完整响应,LLaDA2.1-Flash将其作为可编辑初始化进行修订,实现无需额外训练的异构模型级联
  • 基准测试结果:Fl

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

Research 科学研究 Inference 推理 LLM 大模型 Training 训练