TeaCache4Mochi

TeaCache can speedup Mochi 2x without much visual quality degradation, in a training-free manner. The following video shows the results generated by TeaCache-Mochi with various rel_l1_thresh values: 0 (original), 0.06 (1.5x speedup), 0.09 (2.1x speedup).

https://github.com/user-attachments/assets/29a81380-46b3-414f-a96b-6e3acc71b6c4

📈 Inference Latency Comparisons on a Single A800

mochi-1-preview	TeaCache (0.06)	TeaCache (0.09)
~30 min	~20 min	~14 min

Installation

pip install --upgrade diffusers[torch] transformers protobuf tokenizers sentencepiece imageio

Usage

You can modify the thresh in line 174 to obtain your desired trade-off between latency and visul quality. For single-gpu inference, you can use the following command:

python teacache_mochi.py

Citation

If you find TeaCache is useful in your research or applications, please consider giving us a star 🌟 and citing it by the following BibTeX entry.

@article{liu2024timestep,
  title={Timestep Embedding Tells: It's Time to Cache for Video Diffusion Model},
  author={Liu, Feng and Zhang, Shiwei and Wang, Xiaofeng and Wei, Yujie and Qiu, Haonan and Zhao, Yuzhong and Zhang, Yingya and Ye, Qixiang and Wan, Fang},
  journal={arXiv preprint arXiv:2411.19108},
  year={2024}
}

Acknowledgements

We would like to thank the contributors to the Mochi and Diffusers.