Prefix-Guided Sampling：用短前缀置信度信号优化并行测试时扩展
Early Signals, Strong Decisions: Prefix-Guided Sampling for Parallel Test-Time Scaling
采样前先用短前缀的置信度排个序，只补全有戏的路径，token 省了效果还更好。做推理加速、搞 test-time scaling 的人可以直接看这篇的理论部分和实现思路。
论文提出 Prefix-Guided Sampling（PreG），针对并行测试时扩展（TTS）中低质量推理路径被完整展开后才评估的算力浪费问题给出新解法。方法核心是先生成短前缀，用内部置信度分数对前缀排序，只对最有希望的前缀继续补全，从而把算力重新分配到高质量路径上。论文的理论分析证明，在固定计算预算下，PreG 能通过扩大决策边界间隔来增强有效信号。
作者还提供了一个新视角：置信度信号对接近决策边界的题目影响最大，而经多样本聚合放大后，来自短前缀的小置信度信号就足以影响这些边界情况的结果。实验显示，PreG 在多个基准上一致优于标准并行 TTS 方法，同时显著降低 token 用量和整体计算开销。
Early Signals, Strong Decisions: Prefix-Guided Sampling for Parallel Test-Time Scaling
Parallel test-time scaling (TTS) has been shown to enhance reasoning in large language models by generating and aggregating multiple independent reasoning paths. Internal confidence scores can further improve this process by identifying paths of varying quality. However, existing methods remain computationally inefficient, as low-quality paths are fully expanded before being evaluated. In this work, we aim to improve the use of internal confidence signals for better model performance and computational efficiency. We begin by offering a new perspective on the role of confidence in aggregation, showing that it is most impactful for questions near the decision boundary. Notably, when amplified by multi-sample aggregation, even small confidence signals derived from short prefixes can meaningfully influence outcomes in these boundary cases. Building on this insight, we introduce Prefix-Guided Sampling (PreG), a method that reallocates computation by first generating short prefixes, ranking them using internal confidence scores, and then completing only the most promising candidates. This strategy reduces token usage while maintaining gains in path quality. We provide theoretical analysis demonstrating that, under a fixed compute budget, PreG strengthens the effective signal by widening the decision boundary margin. Empirically, our method consistently outperforms standard parallel TTS approaches across benchmarks, while significantly improving token efficiency and reducing overall computation.
This report is categorized under AITOP's "Papers" desk. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include test-time scaling, 推理模型, Prefix-Guided Sampling, LLM.
Source attribution: curated from arXiv cs.AI (@Jie Ren, Jonathan Rosenfeld, Neil Thompson).
