Created and maintained by OpenAI's Triton team.

Triton is a Python like domain specific language and compiler for writing highly efficient custom GPU kernels.

Hosted on GitHub at openai/triton and distributed via PyPI.

First released in 2019 and now a core dependency of PyTorch 2.x compilation.

It lets researchers write near CUDA performance kernels without hand writing low level CUDA C++.

Used internally by PyTorch torch.compile and depends on LLVM for code generation.

MIT licensed, open source, maintained by OpenAI, roughly 20k GitHub stars, distinct from NVIDIA Triton Inference Server.