Oryx-ViT

Model Summary

The Oryx-ViT model is trained on 200M data and can seamlessly and efficiently process visual inputs with arbitrary spatial sizes and temporal lengths. It is described in the paper Oryx MLLM: On-Demand Spatial-Temporal Understanding at Arbitrary Resolution.

• Repository: https://github.com/Oryx-mllm/Oryx
• Project Page: https://oryx-mllm.github.io
• Languages: English, Chinese

Model Architecture

• Architecture: SigLip
• Data: a mixture of 200M data, 2 epoch
• Precision: BFloat16

Hardware & Software

• Hardware: 64 * NVIDIA Tesla A100
• Orchestration: HuggingFace Trainer
• Code: Pytorch

Citation

@article{liu2024oryx,
title={Oryx MLLM: On-Demand Spatial-Temporal Understanding at Arbitrary Resolution},
author={Liu, Zuyan and Dong, Yuhao and Liu, Ziwei and Hu, Winston and Lu, Jiwen and Rao, Yongming},
journal={arXiv preprint arXiv:2409.12961},
year={2024}
}