Skill Weaving: Efficient LLM Improvement via Modular Skillpacks
摘要
Large language models increasingly require specialization across diverse domains, yet existing approaches struggle to balance multi-domain capacities with strict memory and inference constraints. In this work, we introduce SkillWeave, a modular improvement framework that enables LLMs to specialize under fixed memory budgets. SkillWeave partitions full capabilities of a general-purpose model into skillpacks -- lightweight, domain-specific delta modules -- that reorganize and refine the model's internal knowledge. For efficient deployment, SkillWeave integrates SkillZip to compress skillpacks into compact and inference-ready format, enabling strong multi-domain performance with low-latency execution. On multi-task and agentic benchmarks, a 9B SkillWeave model outperforms several baselines and even surpasses a 32B monolithic LLM, while achieving up to 4x speedup.
引用
@article{arxiv.2605.22205,
title = {Skill Weaving: Efficient LLM Improvement via Modular Skillpacks},
author = {Zhuo Li and Guodong Du and Zesheng Shi and Weiyang Guo and Weijun Yao and Yuan Zhou and Jiabo Zhang and Jing Li},
journal= {arXiv preprint arXiv:2605.22205},
year = {2026}
}
备注
Accepted by ACL2026