arXiv cs.CL· Rui Shi, Xunjian Yin, Jiacheng Sang, Zichuan Liu, Shuyan Zhou·· 12 小时前AI 评分34
WeAct：将动态权重切换作为智能体动作
WeAct: Dynamic Weight Switching as Agent Actions
AI 导读
WeAct 将能力专属的 LoRA 适配器选择加入智能体动作空间，通过模型原生工具调用接口按需组合能力。在八个深度研究基准上，Qwen3-32B 配合 WeAct 平均准确率较全量微调提升 6.2%，较最强静态 LoRA 合并方法提升 27.5%，且计算开销更低。该方法还支持模块化持续适配，新增代码计算适配器或更新写作适配器时主干与其他适配器保持不变。
正文 · 原文
Abstract:Long-horizon agents must learn heterogeneous capabilities from interleaved trajectories. Updating shared parameters to learn new capabilities can degrade previously acquired ones, resulting in catastrophic forgetting. As a trajectory unfolds, different stages may require different capabilities, yet conventional joint training and static adapter merging maintain a fixed parameter configuration throughout execution. We introduce WeAct (Weight-level Internal Actions), which trains independent capability-specific LoRA adapters over a frozen backbone and adds adapter selection to the agent's action space. These internal actions use the model's native tool-calling interface for on-demand capability composition in a shared context. Across eight deep-research benchmarks, Qwen3-32B with WeAct improves average accuracy by 6.2% relative to full fine-tuning counterparts despite using less compute, and by 27.5% relative to the strongest static LoRA-merging method. WeAct also supports modular continual adaptation, by adding a new code computation adapter to expand the agent's capabilities and updating the existing writing adapter to improve report quality. Both updates leave the backbone and other adapters unchanged, supporting modular continual adaptation in agents.
| Comments: | 24 pages, 14 figures, 12 tables, including references and appendices |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.26847 [cs.LG] |
| (or arXiv:2610.26847v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.26847 arXiv-issued DOI via DataCite |
Submission history
From: Rui Shi [view email]
[v1]
Mon, 19 Oct 2026 17:59:19 UTC (1,942 KB)
[v2]
Tue, 20 Oct 2026 04:16:18 UTC (1,943 KB)
来源：arXiv cs.CL · arxiv.org