learnweak-evocua-8b-lora-r32-vscode

LearnWeak: GIMP Adapter for EvoCUA-8B

This repository contains a LoRA adapter for meituan/EvoCUA-8B-20260105 specialized for the GIMP desktop domain. It was developed using LearnWeak, an annotation-free specialization framework for small computer-use agents (CUAs).

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

Model Description

LearnWeak identifies a student agent's weaknesses in a specific software domain using a stronger reference agent (teacher). It then synthesizes targeted tasks and constructs automated supervision. LearnWeak further introduces an error-aware specialization objective that disentangles planning and execution errors, enabling more behaviorally precise updates than broad uniform supervision.

This specific adapter improves the performance of the 8B student agent on tasks related to the GIMP image editor.

How to Get Started with the Model

Serve with vLLM

You can use this adapter with vLLM by enabling LoRA support and providing the module name:

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, such as learnweak-gimp, when calling the served model.

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.