NUS-HPC-AI-Lab/VideoSys
VideoSys is an open-source system for efficient video generation using diffusion transformer models with acceleration techniques like PAB and DSP.
Not currently ranked — collecting fresh signals.
star history
VideoSys provides an easy-to-use system for video generation based on diffusion transformers. It supports multiple open-source video generation models including CogVideoX, Vchitect-2.0, Open-Sora-Plan, and Latte. The system implements acceleration techniques such as Pyramid Attention Broadcast (PAB) for real-time generation and Dynamic Sequence Parallel (DSP) for efficient training of variable-length sequences. The Data-Centric Parallel (DCP) method is also proposed for efficient training of variable sequences.
Frequently asked
- What is NUS-HPC-AI-Lab/VideoSys?
- VideoSys is an open-source system for efficient video generation using diffusion transformer models with acceleration techniques like PAB and DSP.
- Is VideoSys open source?
- Yes — NUS-HPC-AI-Lab/VideoSys is open source, released under the Apache-2.0 license.
- What language is VideoSys written in?
- NUS-HPC-AI-Lab/VideoSys is primarily written in Python.
- How popular is VideoSys?
- NUS-HPC-AI-Lab/VideoSys has 2k stars on GitHub.
- Where can I find VideoSys?
- NUS-HPC-AI-Lab/VideoSys is on GitHub at https://github.com/NUS-HPC-AI-Lab/VideoSys.