Skip to content
Branch: master
Find file Copy path
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
37 lines (23 sloc) 1.92 KB


This is the development repository of Triton, a language and compiler for writing highly efficient custom Deep-Learning primitives.

The formal foundations of this project are described in the following MAPL2019 publication: Triton: An Intermediate Language and Compiler for Tiled Neural Network Computations. Please cite us if you use our work!

The main features of Triton at the moment are:

  • PyTriton: A Python API for writing custom operations for Triton-C compute-kernels. PyTriton automatically generates and just-in-time Tensorflow and PyTorch bindings.
  • Triton-C: An imperative, single-threaded language for writing highly efficient compute-kernels at a relatively high abstraction level using numpy-like extensions of the C language.
  • Triton-IR: An intermediate-representation for optimizing multi-dimensional array operations in linear algebra programs
  • Triton-JIT: An optimizing just-in-time compiler for Triton-C, which generates GPU code on par with state-of-the-art CUDA-C (e.g., CUTLASS) and PTX (e.g., ISAAC). This includes transparent support for mixed-precision and Tensor Cores.


Triton is a fairly self-contained package and uses its own parser (forked from wgtcc) and LLVM code-generator. However, at the moment it still relies on LLVM-8.0+ for PTX code generation.

sudo apt-get install llvm-8-dev
git clone;
cd triton/python/;
python develop;
cd examples;


You can’t perform that action at this time.