learnweak-evocua-8b-lora-r32-thunderbird

learnweak-evocua-8b-lora-r32-gimp

This repository contains a LoRA adapter for meituan/EvoCUA-8B-20260105 specialized for the GIMP software domain using the LearnWeak framework.

LearnWeak is an annotation-free specialization framework for small computer-use agents (CUAs) introduced in the paper Learn from Weaknesses: Automated Domain Specialization for Small Computer-Use Agents. It identifies a student model's weaknesses in a target domain using a stronger reference agent (teacher), synthesizes targeted tasks, and constructs automated supervision to improve performance.

• Project Page: https://learnweak.github.io/
• Repository: https://github.com/sujiikim/LearnWeak
• Paper: https://huggingface.co/papers/2605.28775

Usage

Serve with vLLM

You can serve this adapter alongside the base model using vLLM by enabling LoRA support:

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's API.

Training Details

LearnWeak specializes small computer-use agents for target desktop domains by identifying student weaknesses, generating targeted practice tasks, and training from teacher/student trajectory differences. This specific checkpoint focuses on the GIMP domain.

Citation

If you find this work useful, please consider citing:

@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.