DMax让扩散语言模型更高效地并行生成，错误少、速度更快。
DMax: Aggressive Parallel Decoding for dLLMs
- 用嵌入空间的渐进自修正代替传统掩码转词元，减少错误积累。
- 在GSM8K上推理准确率从2.04提升至5.47，MBPP上从2.71升至5.86。
- 适合追求高速高质文本生成的研究者和工程师使用。
我们提出DMax，一种面向高效扩散语言模型（dLLMs）的新范式。它缓解了并行解码中的误差累积问题，实现激进的并行解码同时保持生成质量。不同于传统掩码dLLMs通过二值掩码到词元的转换进行解码，DMax将解码重构为从掩码嵌入到词元嵌入的渐进自修正过程。核心是「在策略均匀训练」，一种新型训练策略，能高效统一掩码与均匀dLLMs，使模型可从掩码输入及自身错误预测中恢复干净词元。在此基础上，我们提出软并行解码：将每个中间解码状态表示为预测词元嵌入与掩码嵌入的插值，实现嵌入空间内的迭代自修正。大量实验表明其有效性：相比原始LLaDA-2.0-mini，在GSM8K上TPF从2.04升至5.47，保持准确率；在MBPP上，TPF从2.71升至5.86，性能相当。在两块H200 GPU上，批量大小为1时平均达到1,338 TPS。代码已开源：https://github.com/czg1225/DMax。
原文摘要 · Abstract (English)
We present DMax, a new paradigm for efficient diffusion language models (dLLMs). It mitigates error accumulation in parallel decoding, enabling aggressive decoding parallelism while preserving generation quality. Unlike conventional masked dLLMs that decode through a binary mask-to-token transition, DMax reformulates decoding as a progressive self-refinement from mask embeddings to token embeddings. At the core of our approach is On-Policy Uniform Training, a novel training strategy that efficiently unifies masked and uniform dLLMs, equipping the model to recover clean tokens from both masked inputs and its own erroneous predictions. Building on this foundation, we further propose Soft Parallel Decoding. We represent each intermediate decoding state as an interpolation between the predicted token embedding and the mask embedding, enabling iterative self-revising in embedding space. Extensive experiments across a variety of benchmarks demonstrate the effectiveness of DMax. Compared with the original LLaDA-2.0-mini, our method improves TPF on GSM8K from 2.04 to 5.47 while preserving accuracy. On MBPP, it increases TPF from 2.71 to 5.86 while maintaining comparable performance. On two H200 GPUs, our model achieves an average of 1,338 TPS at batch size 1. Code is available at: https://github.com/czg1225/DMax
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。