# TeaCache4LuminaT2X [TeaCache](https://github.com/LiewFeng/TeaCache) can speedup [Lumina-T2X](https://github.com/Alpha-VLLM/Lumina-T2X) 2x without much visual quality degradation, in a training-free manner. The following image shows the results generated by TeaCache-Lumina-Next with various rel_l1_thresh values: 0 (original), 0.2 (1.5x speedup), 0.3 (1.9x speedup), 0.4 (2.4x speedup), and 0.5 (2.8x speedup).  ## 📈 Inference Latency Comparisons on a Single A800 | Lumina-Next-SFT | TeaCache (0.2) | TeaCache (0.3) | TeaCache (0.4) | TeaCache (0.5) | |:-------------------------:|:---------------------------:|:--------------------:|:---------------------:|:---------------------:| | ~17 s | ~11 s | ~9 s | ~7 s | ~6 s | ## Installation ```shell pip install --upgrade diffusers[torch] transformers protobuf tokenizers sentencepiece pip install flash-attn --no-build-isolation ``` ## Usage You can modify the thresh in line 113 to obtain your desired trade-off between latency and visul quality. For single-gpu inference, you can use the following command: ```bash python teacache_lumina_next.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 [Lumina-T2X](https://github.com/Alpha-VLLM/Lumina-T2X) and [Diffusers](https://github.com/huggingface/diffusers).