提出首个路径剪枝系统分类，让大模型推理更快更准
Cut Your Losses! Learning to Prune Paths Early for Efficient Parallel Reasoning
- 按信号来源与可学习性分类路径剪枝方法，发现可学习内部信号潜力
- 新方法STOP在1.5B到20B模型上显著提升效率，20B模型准确率提升至近90%
- 提供可落地的实操指南，适合追求高效推理的部署团队使用
并行推理虽能提升大模型推理能力，但早期错误导致大量无效路径，带来高昂开销。现有路径剪枝研究分散，缺乏统一框架。本文首次构建路径剪枝系统分类，按信号来源（内部/外部）和可学习性（可学习/不可学习）划分。该分类揭示可学习内部方法的巨大潜力，进而提出STOP（Super TOken for Pruning）。在1.5B至20B参数量的大模型上评估显示，STOP相较基线显著提升效果与效率。进一步验证其在不同算力预算下的可扩展性：例如在固定算力下，将GPT-OSS-20B在AIME25上的准确率从84%提升至近90%。最后，提炼出可指导实际部署的实证准则。代码、数据与模型已公开于https://bijiaxihh.github.io/STOP
原文摘要 · Abstract (English)
Parallel reasoning enhances Large Reasoning Models (LRMs) but incurs prohibitive costs due to futile paths caused by early errors. To mitigate this, path pruning at the prefix level is essential, yet existing research remains fragmented without a standardized framework. In this work, we propose the first systematic taxonomy of path pruning, categorizing methods by their signal source (internal vs. external) and learnability (learnable vs. non-learnable). This classification reveals the unexplored potential of learnable internal methods, motivating our proposal of STOP (Super TOken for Pruning). Extensive evaluations across LRMs ranging from 1.5B to 20B parameters demonstrate that STOP achieves superior effectiveness and efficiency compared to existing baselines. Furthermore, we rigorously validate the scalability of STOP under varying compute budgets - for instance, boosting GPT-OSS-20B accuracy on AIME25 from 84% to nearly 90% under fixed compute budgets. Finally, we distill our findings into formalized empirical guidelines to facilitate optimal real-world deployment. Code, data and models are available at https://bijiaxihh.github.io/STOP
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。