Improve model card: Update pipeline tag, add library name, and enrich content for Video-MTR

This PR significantly improves the model card for the `Video-MTR` model by:

* **Updating the `pipeline_tag`**: Changed from `visual-question-answering` to `video-text-to-text`. This new tag more accurately reflects the model's capabilities in long video understanding and multi-turn reasoning for question comprehension, enhancing its discoverability on the Hub (https://huggingface.co/models?pipeline_tag=video-text-to-text).
* **Adding `library_name: transformers`**: Evidence from `config.json` (e.g., `architectures: ["Qwen2_5_VLForConditionalGeneration"]`, `transformers_version: "4.49.0"`) confirms compatibility with the Hugging Face `transformers` library, enabling automated "how to use" code snippets for users.
* **Expanding the model card content**:
* Added the full paper title as the main heading.
* Included the complete paper abstract to provide detailed insights into the model's methodology and contributions.
* The paper link in the content remains `https://arxiv.org/abs/2508.20478` as per instructions.

Please note that the provided GitHub repository and project page URLs were found to be for a different project ("UniMuMo") and have therefore been omitted from this model card to maintain accuracy. No sample usage was included as no relevant code snippets were found.

README.md

@@ -1,11 +1,20 @@
 ---
-license: apache-2.0
-language:
-- en
 base_model:
 - Qwen/Qwen2.5-VL-7B-Instruct
-pipeline_tag: visual-question-answering
+language:
+- en
+license: apache-2.0
+pipeline_tag: video-text-to-text
+library_name: transformers
 ---
+
+# Video-MTR: Reinforced Multi-Turn Reasoning for Long Video Understanding
+
+This model is a checkpoint for **Video-MTR**, presented in the paper [Video-MTR: Reinforced Multi-Turn Reasoning for Long Video Understanding](https://arxiv.org/abs/2508.20478).
+
+## Abstract
+Long-form video understanding, characterized by long-range temporal dependencies and multiple events, remains a challenge. Existing methods often rely on static reasoning or external visual-language models (VLMs), which face issues like complexity and sub-optimal performance due to the lack of end-to-end training. In this paper, we propose Video-MTR, a reinforced multi-turn reasoning framework designed to enable iterative key video segment selection and question comprehension. Unlike traditional video reasoning pipeline, which generate predictions in a single turn, Video-MTR performs reasoning in multiple turns, selecting video segments progressively based on the evolving understanding of previously processed segments and the current question. This iterative process allows for a more refined and contextually aware analysis of the video. To ensure intermediate reasoning process, we introduce a novel gated bi-level reward system, combining trajectory-level rewards based on answer correctness and turn-level rewards emphasizing frame-query relevance. This system optimizes both video segment selection and question comprehension, eliminating the need for external VLMs and allowing end-to-end training. Extensive experiments on benchmarks like VideoMME, MLVU, and EgoSchema demonstrate that Video-MTR outperforms existing methods in both accuracy and efficiency, advancing the state-of-the-art in long video understanding.
+
 ## References
 
-* [Model Paper](https://arxiv.org/abs/2508.20478)
+*   [Model Paper](https://arxiv.org/abs/2508.20478)