让文本生成模型推理速度提升64倍，还保持原质量。
IDLM: Inverse-distilled Diffusion Language Models
- 用逆蒸馏方法在离散空间优化扩散语言模型，解决采样慢问题。
- 实验显示推理步数减少4至64倍，生成质量与教师模型一致。
- 提出稳定梯度的松弛机制，实现离散空间高效训练。
扩散语言模型（DLMs）在文本生成上表现优异，但多步采样导致推理缓慢，限制实际应用。为解决此问题，我们将原本用于连续扩散模型的逆蒸馏技术扩展至离散场景。然而，该扩展带来理论与实践双重挑战：理论上，逆蒸馏目标缺乏唯一性保障，可能产生次优解；实践中，离散空间反向传播难以实现且易不稳定。为此，我们首先证明所提逆形式具有唯一解，确保优化有效性；随后引入梯度稳定的松弛策略以支持有效训练。实验表明，在多个DLM模型上，所提出的逆蒸馏扩散语言模型（IDLM）可将推理步数减少4至64倍，同时保持教师模型的生成质量。代码、模型检查点及视频教程已公开于项目页面：https://david-cripto.github.io/idlm-project-page。
原文摘要 · Abstract (English)
Diffusion Language Models (DLMs) have recently achieved strong results in text generation. However, their multi-step sampling leads to slow inference, limiting practical use. To address this, we extend Inverse Distillation, a technique originally developed to accelerate continuous diffusion models, to the discrete setting. Nonetheless, this extension introduces both theoretical and practical challenges. From a theoretical perspective, the inverse distillation objective lacks uniqueness guarantees, which may lead to suboptimal solutions. From a practical standpoint, backpropagation in the discrete space is non-trivial and often unstable. To overcome these challenges, we first provide a theoretical result demonstrating that our inverse formulation admits a unique solution, thereby ensuring valid optimization. We then introduce gradient-stable relaxations to support effective training. As a result, experiments on multiple DLMs show that our method, Inverse-distilled Diffusion Language Models (IDLM), reduces the number of inference steps by 4x-64x, while preserving the teacher model's generation quality. We provide the code, model checkpoints, and video tutorials on the project page: https://david-cripto.github.io/idlm-project-page
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。