---
id: 20260821-T0-06
title: "新方法SLAaaT：让智能体动态切换LoRA适配器，避免灾难性遗忘"
title_en: "SLAaaT Lets Agents Switch LoRA Adapters On-the-Fly, Preventing Catastrophic Forgetting"
url: https://ai.daily.yangsir.net/daily/20260821-T0-06
issue_date: 2026-08-21
publish_date: 2026-08-20T04:00:00.000Z
category: research
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2608.17034
---

# 新方法SLAaaT：让智能体动态切换LoRA适配器，避免灾难性遗忘

来自arXiv的新研究提出SLAaaT（Switching LoRA Adapters as a Tool），让智能体在长任务执行中动态调用不同LoRA适配器，按需切换能力。传统后训练虽能提升专项任务表现，但常导致其他领域能力灾难性遗忘。SLAaaT将适配器视为“工具”，智能体可自主选择组合，从而在复杂轨迹中兼顾多项能力，无需重新训练主模型。

## English Version

**SLAaaT Lets Agents Switch LoRA Adapters On-the-Fly, Preventing Catastrophic Forgetting**

New research on arXiv introduces SLAaaT (Switching LoRA Adapters as a Tool), enabling agents to dynamically invoke different LoRA adapters during long trajectories. While post-training boosts specialized task performance, it often causes catastrophic forgetting elsewhere. SLAaaT treats adapters as tools agents can select and compose, letting them balance multiple capabilities in complex missions without retraining the base model.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2608.17034)

**详情页**：https://ai.daily.yangsir.net/daily/20260821-T0-06

---

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