#3232nd of 37 in ML frameworks and libraries
triton
Python-based language and compiler for writing custom GPU kernels
Documentation ↗
(opens in a new tab)Website ↗
(opens in a new tab)Repository on GitHub ↗
(opens in a new tab)
- Stars
- 20.3k
- License
- MIT
- Last commit
- Oct 2026
- Last release
- Aug 2026
- Language
- MLIR
Missed the latest GitHub check; these facts may be older.
Overview
Triton is a Python DSL and MLIR/LLVM-based compiler for writing tiled deep-learning kernels that compile to NVIDIA and AMD GPU code. It aims for higher productivity than CUDA and more flexibility than other DSLs. It installs via pip, with wheels for CPython 3.10-3.14, and includes a CPU interpreter mode for debugging.
Who it is for: GPU kernel authors and ML compiler engineers writing custom DL primitives
Strengths
- Kernels are written in Python and compiled through MLIR and LLVM to PTX or AMDGCN
- TRITON_INTERPRET=1 runs kernels without a GPU, allowing Python breakpoints
- Many debug knobs: per-pass MLIR IR dumps, kernel IR override, reproducer files, autotune printing
- Targets both NVIDIA (compute capability 8.0+) and AMD (ROCm 6.2+) GPUs
Weaknesses
- Linux only; CPU backend is listed as under development
- Building from source needs a matching LLVM version; LLVM has no stable API
- No single command to run all tests; running them needs a GPU or the nogpu subset
- Aimed at kernel authors, not a ready-made model runner
What it needs
- GPU required
- Needs LLVM, PyTorch (for tests), CUDA or ROCm
Also in ML frameworks and libraries
See all 37| Rank | Project | Score |
|---|---|---|
| 1 | transformersPython library defining open models for text, vision, audio and multimodal tasks | 95 out of 100 |
| 2 | pytorchPython tensor library with GPU support and tape-based autograd for deep learning | 94 out of 100 |
| 3 | ultralyticsPython library and CLI for training and running YOLO vision models | 90 out of 100 |
| #4 | supervisionPython toolkit for loading, annotating and counting computer vision detections | 87 out of 100 |
| #5 | pytorch-image-modelsPyTorch library of image models, pretrained weights and reference training scripts | 86 out of 100 |