learnweak-evocua-8b-lora-r32-os

LearnWeak: EvoCUA-8B GIMP Adapter

This repository contains a LoRA adapter for meituan/EvoCUA-8B-20260105 specialized for the GIMP (GNU Image Manipulation Program) domain. It was developed using the LearnWeak framework.

• Paper: Learn from Weaknesses: Automated Domain Specialization for Small Computer-Use Agents
• Project Page: https://learnweak.github.io/
• Repository: https://github.com/sujiikim/LearnWeak

Model Description

LearnWeak is an annotation-free specialization framework for small computer-use agents (CUAs). It uses a stronger reference agent to identify a student model's weaknesses in a target domain, synthesizes targeted tasks, and constructs automated supervision. This specific adapter improves the agent's performance in the GIMP software domain by training on teacher/student trajectory differences.

On OSWorld, LearnWeak achieves significant performance gains (average 11.1-11.6 percentage points) over base models like EvoCUA-8B and OpenCUA-7B across multiple software domains.

How to Get Started with the Model

Serve with vLLM

You can serve the base model with this LoRA adapter enabled using vLLM:

vllm serve meituan/EvoCUA-8B-20260105 \
  --enable-lora \
  --max-lora-rank 32 \
  --lora-modules learnweak-gimp=SujiKim/learnweak-evocua-8b-lora-r32-gimp

Use the LoRA module name learnweak-gimp when calling the served model API.

Citation

@article{kim2026learnweaknessesautomateddomain,
  title   = {Learn from Weaknesses: Automated Domain Specialization for Small Computer-Use Agents},
  author  = {Kim, Suji and Kim, Kangsa and Hwang, Sung Ju},
  journal = {arXiv preprint arXiv:2605.28775},
  year    = {2026}
}

Acknowledgments

This project builds on OSWorld, LlamaFactory, and EvoCUA.