Skill-Uni jointly trains LLM skill retrieval and use
Original title: Two Birds with One Stone: Skill-Augmented Generation via Unified Retrieval and Utilization
Authors report average gains of 66.59% on Qwen3-4B and 34.82% on LLaMA3.1-8B across five benchmarks.
Skill-Uni is a reinforcement-learning framework that uses one model for generative skill retrieval and task solving, rather than separate models for each step. Training combines warm-up instruction tuning with GRPO. The authors report average gains of 66.59% over base Qwen3-4B and 34.82% over LLaMA3.1-8B on five skill-related benchmarks, and say the method works with multiple retrieval engines.
Unconfirmed: Benchmark gains are author-reported and not independently verified here.
- Existing methods split skill retrieval and utilization across separate models.
- Training uses warm-up instruction tuning followed by GRPO.
- Reported average gains are 66.59% and 34.82% on the two base models.
- Authors say Skill-Uni is compatible with various skill retrieval engines.
- AI models
- Qwen3-4BLLaMA3.1-8B
From the original article
1 min read at arxiv.orgSkills, as reusable external knowledge, have demonstrated substantial potential to enhance the ability of Large Language Models (LLMs) to solve complex tasks. However, existing skill-augmented generation methods typically rely on separate models for skill retrieval and skill utilization, resulting in disconnected optimization of these two processes.
By Zihan Luo · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.