Future-L1-SFT

Future-L1-SFT is the SFT checkpoint used by Future-L1: Imagine Before You Predict, a framework for video event prediction with interleaved language and latent visual reasoning.

Project repository: https://github.com/OpenGVLab/Future-L1

Paper: https://arxiv.org/abs/2606.05769

View on alphaXiv: https://www.alphaxiv.org/abs/2606.05769

Training set: Future-L1-50K

Checkpoint files: model-00001-of-00004.safetensors, model-00002-of-00004.safetensors, model-00003-of-00004.safetensors, and model-00004-of-00004.safetensors

Model configuration: SwimBird_Qwen3VL with qwen3_vl model type

Citation

If you use this checkpoint, please cite the paper:

@article{jiang2026imagine,
  title={Imagine Before You Predict: Interleaved Latent Visual Reasoning for Video Event Prediction},
  author={Jiang, Tianxiang and Wu, Linquan and Xia, Sheng and Li, Songze and Yan, Ziang and Yang, Haoyu and Qiao, Yu and Wang, Yi},
  journal={arXiv preprint arXiv:2606.05769},
  year={2026}
}