Skip to content

AITemplate is a Python framework which renders neural network into high performance CUDA/HIP C++ code. Specialized for FP16 TensorCore (NVIDIA GPU) and MatrixCore (AMD GPU) inference.

License

Notifications You must be signed in to change notification settings

hlky/AITemplate

Repository files navigation

hlky's AITemplate (HAIT)

HAIT continues development from facebookincubator/AITemplate.

Major changes include:

Installation

Requirements

Install CUDA Toolkit

For Windows, install Build Tools for Visual Studio 2022 or Visual Studio 2022 (Community, Professional or Enterprise).

If Build Tools or Visual Studio is installed after CUDA Toolkit, re-run CUDA Toolkit installation to get Visual Studio integration. Build Tools only may need CUDA X.Y.props copying manually, refer to CUDA documentation for the default .props install directory and adjust for your Build Tools install directory.

Python

git clone --recursive https://github.com/hlky/AITemplate
cd AITemplate/python
pip install -e .

Build release

python setup.py bdist_wheel
pip uninstall -y aitemplate
pip install dist/*.whl

Roadmap

Generally:

  • More kernels
  • More modeling support

Refer to Issues and Projects.

Acknowledgements

With thanks to the original developers and other Meta engineers.

License

AITemplate is licensed under the Apache 2.0 License.

About

AITemplate is a Python framework which renders neural network into high performance CUDA/HIP C++ code. Specialized for FP16 TensorCore (NVIDIA GPU) and MatrixCore (AMD GPU) inference.

Resources

License

Code of conduct

Stars

Watchers

Forks

Packages

No packages published