arXiv cs.AI· Ya-Shi Zhang, Jian Tang, Pierre-Andr\'e No\"el, Oleksiy Ostapenko·· 2 天前AI 评分27
TIPS：面向扩散语言模型扩展的微型推理时策略
TIPS: A Tiny Inference-time Policy for Scaling in Diffusion Language Models
AI 导读
研究者提出 TIPS，一个约 600K 参数的推理时控制器，让 LLaDA-8B-Instruct 在解码时同时学习取消掩码与重新掩码，而底层语言模型保持冻结。
正文 · 原文
Abstract:Masked diffusion language models expose an inference-time choice: which positions become visible after each denoising pass. Many existing decoding methods use hand-designed rules or require visibility to grow monotonically. We present a $\textbf{T}$iny $\textbf{I}$nference-time $\textbf{P}$olicy for $\textbf{S}$caling ($\textbf{TIPS}$), an approximately $600$K-parameter controller that learns both unmasking and remasking while keeping the underlying language model frozen. TIPS uses visibility-aware confidence features and a shallow transformer to select which tokens should be visible, while a fixed schedule controls the expected visible fraction. We train the controller through supervised warmup followed by reinforcement learning from terminal correctness rewards. On GSM8K and MATH-500 with LLaDA-8B-Instruct, the full-canvas and two-block TIPS variants achieve $58.46\%$ and $57.97\%$ micro-averaged accuracy at a budget of $32$ decoder evaluations, exceeding the strongest baseline in our main comparison by $12.1$ and $11.6$ percentage points, respectively. Both variants are trained at a budget of $32$ evaluations and remain competitive at higher budgets without retraining. These results demonstrate that learning when to reveal and reconsider tokens can improve the accuracy-compute trade-off without updating the decoder backbone.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.22099 [cs.LG] |
| (or arXiv:2610.22099v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.22099 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Ya Shi Zhang [view email]
[v1]
Thu, 15 Oct 2026 15:46:55 UTC (926 KB)
来源：arXiv cs.AI · arxiv.org