让扩散语言模型更精准执行任意位置的指令约束。
In-Place Instruction Following in Diffusion Language Models
- 提出在生成时动态调整注意力，强化对指令位置的响应。
- 在四个模型上平均得分提升15.35点，关键约束任务提升超15点。
- 适合需要精准控制生成内容位置和风格的研究与应用。
扩散大语言模型通过双向迭代去噪生成文本，天然支持在任意输出位置设置用户约束，这一范式称为就地提示（IPP）。本文将此形式化为就地指令遵循（IIF）任务，并构建了涵盖字面、风格和语用功能约束的分层基准IIF-Bench，配套基于评分的局部-全局评估协议。推理阶段的注意力偏置探测表明，原始dLLMs在去噪过程中常低估约束片段的重要性。为此，提出GRAFT框架，结合约束感知SFT与偏好优化的后训练策略。在四个代表性dLLMs上，GRAFT将平均IIF得分从57.75提升至73.10（+15.35），在字面和语用功能约束上分别获得15.91和15.57的绝对提升，同时保持通用生成能力。
原文摘要 · Abstract (English)
Diffusion Large Language Models (dLLMs) generate text via bidirectional iterative denoising, naturally supporting user-specified constraints anchored at arbitrary output positions, a paradigm known as In-place Prompting (IPP). We formalize this as the In-place Instruction Following (IIF) task and construct IIF-Bench, a hierarchical benchmark spanning literal, style, and discourse-function constraints, paired with a rubric-based local-global evaluation protocol. An inference-time attention-bias probe suggests that vanilla dLLMs often under-prioritize constraint spans during denoising. We then propose GRAFT, an IPP-oriented post-training framework combining constraint-aware SFT and preference optimization. On four representative dLLMs, GRAFT raises the average IIF score from 57.75 to 73.10 (+15.35 points), with absolute gains of 15.91 and 15.57 points on literal and discourse-function constraints, while preserving general generation ability.
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。