chengzeyi/ParaAttention
Context-parallel attention library that accelerates diffusion transformer model inference with dynamic caching.
Not currently ranked — collecting fresh signals.
star history
ParaAttention implements Ulysses and Ring style parallelism for attention computation to speed up DiT model inference. It supports dynamic caching strategies and integrates with popular diffusion libraries including Hugging Face diffusers. The library specifically targets image and video generation models such as FLUX-dev and Wan 2.1 for text-to-image, text-to-video, and image-to-video tasks.
Frequently asked
- What is chengzeyi/ParaAttention?
- Context-parallel attention library that accelerates diffusion transformer model inference with dynamic caching.
- Is ParaAttention open source?
- Yes — chengzeyi/ParaAttention is an open-source project tracked on heatdrop.
- What language is ParaAttention written in?
- chengzeyi/ParaAttention is primarily written in Python.
- How popular is ParaAttention?
- chengzeyi/ParaAttention has 428 stars on GitHub.
- Where can I find ParaAttention?
- chengzeyi/ParaAttention is on GitHub at https://github.com/chengzeyi/ParaAttention.