论文/arXiv 2610.09074·22 天前

TAP：通过轨迹锚定恢复实现高效长时程智能体剪枝

TAP面向长时程智能体推理成本，结合结构剪枝与轨迹锚定在线恢复，并在ALFWorld和WebShop上验证压缩效果。

AI 整理

• 本体 SDK 与控制
• 具身模型
• 端侧部署

原文链接：TAP: Efficient Long-Horizon Agent Pruning via Trajectory-Anchored Recovery（arXiv 2610.09074 · 发布日期：2026-10-06）

TAP: Efficient Long-Horizon Agent Pruning via Trajectory-Anchored Recovery

TAP面向长时程智能体推理成本，结合结构剪枝与轨迹锚定在线恢复，并在ALFWorld和WebShop上验证压缩效果。

核心要点

• 长时程智能体需反复调用模型，现有问答剪枝方法迁移到智能体会严重降低任务表现。
• TAP将结构剪枝与高效在线恢复结合，用教师轨迹锚定交互并让学生生成推理-动作响应。
• 冻结密集教师监督学生响应前缀，并基于恢复目标梯度对已恢复学生重新评估通道。
• 移除60% FFN通道后，在两个任务上分别保留99.2%和88.0%成功率。

验证范围与局限：验证范围为密集7B智能体在ALFWorld和WebShop上的任务成功率与GPU时间。

英文摘要 (Abstract)

Emerging long-horizon agentic tasks require repeated model calls, worsening the inference cost of already-costly language models. While narrow agentic tasks suggest potential for aggressive model pruning without performance drop, empirical results show existing methods proposed for question answering tasks severely degrade task performance when applied to agentic models. We trace this failure to two decisions: what to prune and how to recover. For pruning, one-shot importance estimates fail to track how the pruned model adapts. For recovery, offline distillation covers only teacher prefixes, while full-trajectory on-policy distillation causes student errors to compound across turns. In this work, we propose Trajectory-Anchored Pruning (TAP), the first structural pruning framework for reinforcement learning (RL)-trained agents. TAP couples structural pruning with efficient on-policy recovery, anchoring interactions to teacher trajectories while allowing the student to generate each reasoning-action response. A frozen dense teacher supervises the student's response prefixes, addressing within-response training-inference mismatch while preventing student-induced deviations from propagating across training turns. Instead of one-shot pruning, TAP re-scores channels using gradients of the recovery objective on the recovered student, connecting iterative channel selection to the evolving policy. With 60% of FFN channels removed, TAP retains 99.2% and 88.0% of the dense 7B agents' task success rates on ALFWorld and WebShop, respectively, while reducing GPU time per successful task by approximately 22% and 17%. These results demonstrate effective structural compression of long-horizon agents under a limited recovery budget.

作者：Yuanzhe Li, Pengxin Wang, Yuxin Ren, Jianing Deng, Jingtong Hu, Song Wang, Jingdi Chen, Huanrui Yang

来源

本页事实对应的公开页面，以原文为准。

1. arXiv 2610.09074：TAP: Efficient Long-Horizon Agent Pruning via Trajectory-Anchored Recovery访问 2026-10-21