README.md

metadata

language:
  - en
license: mit
base_model: xlangai/OpenCUA-7B
tags:
  - world-model
  - computer-use
  - transition-prediction
  - lora
  - peft
library_name: peft

World Model LoRA Adapter

LoRA adapter fine-tuned on ageppert/world-model-transitions for predicting GUI state transitions in desktop computer-use tasks.

Base Model

xlangai/OpenCUA-7B

Usage

from peft import PeftModel
from transformers import AutoModelForVision2Seq, AutoProcessor

base_model = AutoModelForVision2Seq.from_pretrained("xlangai/OpenCUA-7B", trust_remote_code=True)
model = PeftModel.from_pretrained(base_model, "ageppert/world-model-7b-lora")
processor = AutoProcessor.from_pretrained("ageppert/world-model-7b-lora")

Training

• LoRA rank: 16, alpha: 32
• Target modules: ['q_proj', 'k_proj', 'v_proj', 'o_proj']
• Learning rate: 0.0002
• Epochs: 3
• Train loss: 4.954768469485616
• Eval loss: 0.5774359703063965

Citation

Based on OpenCUA (arXiv:2508.09123).