---
license: apache-2.0
language:
- en
base_model:
- Qwen/Qwen2.5-VL-7B-Instruct
pipeline_tag: visual-question-answering
---

# Video-MTR

Video-MTR: Reinforced Multi-Turn Reasoning for Long Video Understanding.

This checkpoint extends Qwen2.5-VL-7B-Instruct with a multi-turn frame-retrieval
policy trained via PPO, with an 80-frame video input budget.

## References

* [Paper](https://arxiv.org/abs/2508.20478)
* [Code](https://github.com/Xyuan13/Video-MTR)