---
id: 20260620-T0-05
title: "因果归因剪枝（CAP）：无需训练即可保留LLM推理性能"
title_en: "Causal Attribution Pruning Preserves LLM Reasoning Without Training"
url: https://ai.daily.yangsir.net/daily/20260620-T0-05
issue_date: 2026-06-20
publish_date: 2026-06-19T04:00:00.000Z
category: research
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2606.19350
---

# 因果归因剪枝（CAP）：无需训练即可保留LLM推理性能

一项新研究提出了因果归因剪枝（CAP），这是一种无需训练的剪枝方法。通过测量注意力头对推理输出的因果影响来识别关键节点，CAP 能够在大幅降低推理成本的同时，保留大模型的多步推理性能。

## English Version

**Causal Attribution Pruning Preserves LLM Reasoning Without Training**

New research introduces Causal Attribution Pruning (CAP), a training-free method that identifies critical attention heads by measuring their causal impact on reasoning. CAP reduces inference costs while preserving multi-step reasoning performance.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2606.19350)

**详情页**：https://ai.daily.yangsir.net/daily/20260620-T0-05

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*