用教师模型轨迹指导学生，实现少步生成高质量文本
Few-Step Diffusion Language Models via Trajectory Self-Distillation
- 用教师模型的生成轨迹监督学生模型，减少分步生成误差
- 少步推理在推理与代码生成任务上接近全步效果
- 适合追求高效生成的场景，如实时对话系统
扩散语言模型（DLLMs）有望通过并行解码实现快速文本生成，但实际中减少解码步数常导致输出质量显著下降，主要源于标记因子化误差。为此，我们提出一种自蒸馏框架，让少步学生模型匹配全步教师模型的生成轨迹。理论与实证均表明，轨迹级监督可缓解因子化误差，从而实现有效少步解码。进一步引入直接判别优化（DDO），采用反KL目标促使学生向教师模式聚焦，在复杂推理任务上表现更优。在多个推理与代码生成基准上，该方法显著缩小了少步与全步解码之间的性能差距。源代码已开源：https://github.com/Tyrion58/T3D。
原文摘要 · Abstract (English)
Diffusion large language models (DLLMs) have emerged as powerful generative models with the promise of fast text generation through parallel decoding. However, realizing this potential in practice remains challenging: reducing the number of decoding steps, typically causes a substantial degradation in output quality due to token factorization error. To alleviate this, we propose a self-distillation framework that trains a few-step student to match the generative trajectory of a full-step teacher. We theoretically and empirically show that trajectory-level supervision mitigates this factorization error, thereby enabling effective few-step decoding. We further incorporate Direct Discriminative Optimization (DDO), a reverse-KL objective that encourages mode-seeking toward the teacher's modes, yielding stronger performance on challenging reasoning tasks. Across reasoning and code-generation benchmarks, our method substantially narrows the gap between few-step and full-step decoding. The source code is available at https://github.com/Tyrion58/T3D.
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。