Skill Weaving: Efficient LLM Improvement via Modular Skillpacks
For LLM practitioners needing efficient multi-domain specialization under strict memory and latency constraints, SkillWeave offers a practical modular framework.
SkillWeave partitions a general-purpose LLM into lightweight domain-specific skillpacks, enabling strong multi-domain performance under fixed memory. A 9B model outperforms several baselines and surpasses a 32B monolithic LLM with up to 4x speedup.
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.