LongVT-Source

This repository contains the source video and image files for the LongVT project.

Overview

LongVT is an end-to-end agentic framework that enables "Thinking with Long Videos" via interleaved Multimodal Chain-of-Tool-Thought. This dataset provides the raw media files referenced by the training annotations in LongVT-Parquet.

Dataset Structure

The source files are organized by dataset type and stored as zip archives:

Training Data

Source	Description	Files
longvideoreason	Long video reasoning data	66 zips
videor1	Video-R1 COT data	13 zips
longvideoreflection	Long video reflection data	27 zips
selftrace	Self-distilled iMCoTT traces	6 zips
tvg	Temporal video grounding data	2 zips
geminicot	Gemini-distilled COT data	2 zips
llavacot	LLaVA COT data	1 zip
openvlthinker	OpenVLThinker data	1 zip
wemath	WeMath data	1 zip
selfqa	Self-curated QA for RL	1 zip
rl_val	RL validation data	1 zip

Evaluation Data

Source	Description	Files
videosiaheval	VideoSIAH-Eval benchmark videos	12 zips

Download

Install huggingface_hub

pip install huggingface_hub

Download all source files

huggingface-cli download longvideotool/LongVT-Source --repo-type dataset --local-dir ./source

Or download specific files

huggingface-cli download longvideotool/LongVT-Source longvideoreason_1.zip --repo-type dataset --local-dir ./source## Usage

After downloading, extract the zip files to obtain the source media:

cd source unzip "*.zip"The extracted paths will match those referenced in the LongVT-Parquet annotations.

Related Resources

• 📄 Paper: arXiv:2511.20785
• 🌐 Project Page: LongVT Website
• 💻 Code: GitHub Repository
• 📊 Annotations: LongVT-Parquet
• 🤗 Models: LongVT Collection

Citation

If you find LongVT useful for your research and applications, please cite using this BibTeX:

@misc{yang2025longvtincentivizingthinkinglong,
      title={LongVT: Incentivizing "Thinking with Long Videos" via Native Tool Calling}, 
      author={Zuhao Yang and Sudong Wang and Kaichen Zhang and Keming Wu and Sicong Leng and Yifan Zhang and Bo Li and Chengwei Qin and Shijian Lu and Xingxuan Li and Lidong Bing},
      year={2025},
      eprint={2511.20785},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2511.20785}, 
}

License

This dataset is released under the Apache 2.0 License.