通过分阶段生成提升扩散语言模型推理速度，最高提速9.1倍。
SAID: Accelerating Diffusion-Based Language Models via Scaffold-Aware Iterative Decoding
- 先聚焦关键词建立骨架结构，再补细节，减少冗余计算。
- 在多个基准上实现最高9.1倍加速，性能仍保持领先。
- 适合追求高效生成的开发者与研究者使用。
扩散式大语言模型（DLLMs）通过双向上下文迭代去噪实现非自回归生成，但高质生成需大量去噪步骤，导致推理成本高昂。本文提出SAID框架，通过分阶段重分配计算资源：首先对关键骨架令牌进行去噪以建立粗粒度语义结构，随后仅对可预测的细节令牌用较少步数完成生成。进一步将SAID扩展至块级扩散解码，并引入置信度分层生成机制（CHLG），仅对低置信度令牌分配额外步骤。在LLaDA-8B和LLaDA 1.5模型上，跨数学、编码与知识类基准测试显示，SAID可实现最大9.1倍的推理加速，同时保持优异性能。代码已开源：https://github.com/TH-AI-Lab-PKU/SAID。
原文摘要 · Abstract (English)
Diffusion large language models (DLLMs) enable non-autoregressive generation by iteratively denoising corrupted token sequences with bidirectional context. Despite their ability to update multiple positions in parallel, inference remains costly due to the many denoising steps required for high-quality generation. We propose SAID, a Scaffold-Aware Iterative Decoding framework that accelerates DLLMs by reallocating computation across tokens. SAID first spends denoising computation on scaffold tokens to establish the coarse semantic structure, and then completes predictable detail tokens with fewer steps. We further adapt SAID to block-wise diffusion decoding and introduce Confidence-Hierarchical Layered Generation (CHLG), which assigns additional steps only to low-confidence tokens. Experiments on LLaDA-8B and LLaDA 1.5 across math, coding, and knowledge benchmarks show that SAID significantly accelerates DLLM inference with a maximum speedup of 9.1x while maintaining competitive performance. Our code is publicly available: https://github.com/TH-AI-Lab-PKU/SAID.
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。