A Fusion Code Generator for NVIDIA GPUs (commonly known as "nvFuser")

We publish nightly wheel packages on https://pypi.nvidia.com, while build against stable torch version on https://pypi.org and https://pypi.nvidia.com. Wheels are published for Python version: 3.10, 3.12.

Note that nvfuser built against stable torch version isn't compatible with nightly pytorch wheel, so ensure you pick the right version suiting your environment.

You can install a nightly nvfuser pip package built against torch nightly code base with pip install --pre nvfuser-cu128 --extra-index-url https://pypi.nvidia.com

As we build against nightly torch wheel and there's no compatibility promised on nightly wheels, we have explicitly marked the nightly torch wheel as an optional dependency. You can choose to install the torch wheel along with nvfuser package, e.g. pip install --pre "nvfuser-cu128[torch]" --extra-index-url https://pypi.nvidia.com. Note that this may uninstall your local pytorch installation and install the compatible nightly pytorch.

Nvfuser pip wheel built against stable torch releases is published on both https://pypi.org and https://pypi.nvidia.com. Pick the right cuda toolkit version to match your torch installation. e.g. pip install nvfuser-cu128-torch27.

For old nvfuser builds against old version pytorch, e.g. nvfuser-cuXXY-torchZW, there are packages available at PyPI. We build and publish builds against the latest stable pytorch on https://pypi.org on 1st and 15th of every month regularly and when major changes are added.

We always recommend use of the latest nvfuser build with latest cuda and pytorch versions.

PyPI: https://pypi.org/project/nvfuser/

Docs: https://github.com/NVIDIA/Fuser/wiki

git clone https://github.com/NVIDIA/Fuser.git
cd Fuser
pip install -r python/requirements.txt

[DEPRECATED] `[MAX_JOBS] python setup.py develop [args]`
pip install --no-build-isolation -e python -v

Prerequisites are automatically validated before building. If validation fails, follow the error message instructions.

Validated: Python, CMake, Ninja, pybind11, PyTorch (CUDA), System CUDA toolkit, Git submodules, GCC, LLVM

Skip validation (for CI or custom setups):

NVFUSER_BUILD_SKIP_VALIDATION=1 pip install --no-build-isolation -e python -v

See python/setup.py for all build environment variables.

Supported compilers:

GCC:

GCC 13+ is required (C++20 <format> header support). Recommended versions:

• gcc 13.3
• gcc 14.2

Note: GCC 12 is not supported due to missing <format> header in libstdc++.

Clang:

• clang 19+

Supported C++ standard:

• C++20