PtRain/AbductiveMLLM

Model Card

Model Details

• Finetuned from model: Qwen2-VL-7B-Instruct
• Finetuned dataset: VAR, YoucookII
• Paper: ABDUCTIVEMLLM: Boosting Visual Abductive Reasoning Within MLLMs

How to Get Started with the Model

Use the code below to get started with the model.

    model = Qwen2VLForConditionalGeneration.from_pretrained(
        'path/to/your/Qwen2-VL-7B-Instruct',
        torch_dtype='auto',
        attn_implementation="flash_attention_2"
    )

    # merge lora adapter from Training Stage-1
    lora_model_dir = 'qwen2vl_7b_var_lora1' 
    model = PeftModel.from_pretrained(model, model_id=lora_model_dir) 
    model = model.merge_and_unload()

    # merge lora adapter from Training Stage-2
    lora2_dir = 'qwen2vl_7b_var_select3_lora2'
    model = PeftModel.from_pretrained(model, model_id=lora2_dir)
    model = model.merge_and_unload()

    model.generate()

Citation

@article{chang2026abductivemllm,
  title={AbductiveMLLM: Boosting Visual Abductive Reasoning Within MLLMs},
  author={Chang, Boyu and Wang, Qi and Guo, Xi and Nan, Zhixiong and Yao, Yazhou and Zhou, Tianfei},
  journal={arXiv preprint arXiv:2601.02771},
  year={2026}
}