license: apache-2.0
library_name: transformers
pipeline_tag: image-text-to-text
base_model: Qwen/Qwen2.5-VL-7B-Instruct
tags:
- visual-reasoning
- tool-use
- iterative-reasoning
- grpo
Dynamic Tool Orchestration for Iterative Visual Reasoning
This repository contains AdaReasoner-TC-7B-Randomized, a variant of the model presented in AdaReasoner: Dynamic Tool Orchestration for Iterative Visual Reasoning.
🔔 Important Note on Model Status
The models released on this page belong to the AdaReasoner-TC series and are not the final RL-fine-tuned models. They are trained using Tool Cold Start (TC) supervised fine-tuning only, and are intended for analysis, ablation, and reproducibility purposes.
For RL fine-tuned version, please refer to Data & models
📋 Model Description
AdaReasoner-7B is a vision-language model trained with dynamic tool orchestration capabilities for iterative visual reasoning.
AdaReasoner-TC series are trained through TC (Tool Cold Start) supervised fine-tuning only, without subsequent RL fine-tuning.
This specific variant, AdaReasoner-TC-7B-Randomized, is trained with the adaptive learning method, enabling strong generalization to unseen tools and tasks. It is designed for open-ended and evolving tool environments where adaptability is required.
Key Differences between TC variants:
- Randomized: Trained with adaptive learning method, enabling zero-shot generalization to novel tools and task configurations.
- Non-Randomized: Trained without adaptive learning, offering more predictable behavior on familiar tools but lacking generalization.
📊 Performance
Please refer to our paper for detailed benchmark results across multiple visual reasoning tasks. AdaReasoner improves the 7B base model by +24.9% on average and surpasses strong proprietary systems such as GPT-5 on multiple tasks, including VSP and Jigsaw.
📚 Citation
If you use this model in your research, please cite:
@article{song2026adareasoner,
title={AdaReasoner: Dynamic Tool Orchestration for Iterative Visual Reasoning},
author={Song, Mingyang and Sun, Haoyu and Gu, Jiawei and Li, Linjie and Xu, Luxin and Krishna, Ranjay and Cheng, Yu},
journal={arXiv preprint arXiv:2601.18631},
year={2026}
}
📄 License
Apache 2.0
🤝 Acknowledgments
This model is part of the AdaReasoner project. For more information, visit our GitHub repository.
📧 Contact
For questions and feedback, please open an issue in our GitHub repository.