Installation with Nvidia GPU

In this section, we provide the installation guide for Nvidia GPU.

VeOmni also supports other hardware platform, please refer to Ascend.

Required Environment

CUDA == 12.8

Install with uv or pip

UV

Recommend to use uv for faster and easier installation.

git clone https://github.com/ByteDance-Seed/VeOmni.git
cd VeOmni

# use the locked uv env
uv sync --locked  --extra gpu
source .venv/bin/activate

You can use --extra to install other optional dependencies. Refer to pyproject.toml for more details.

# eg. install with dit dependencies in GPU
uv sync --locked  --extra gpu --extra dit

# eg. install with video/audio processing dependencies (torchcodec, PyAV, librosa, soundfile)
# Note: `video` and `audio` extras are equivalent - both include video and audio processing
uv sync --locked  --extra gpu --extra video
# or equivalently:
uv sync --locked  --extra gpu --extra audio

Note: For video/audio processing with the video or audio extra, you also need to install ffmpeg separately:

# Ubuntu/Debian
sudo apt-get install ffmpeg

# macOS
brew install ffmpeg

Pip

git clone https://github.com/ByteDance-Seed/VeOmni.git
cd VeOmni

pip3 install -e .[gpu]