Triton
Write custom deep learning GPU kernels with a Python-based language and compiler for Linux, supporting NVIDIA and AMD hardware.
Triton is a programming language and compiler for developers writing custom GPU computations for deep learning. Its Python-based environment provides control over the kernels behind model operations.
- Custom kernels: Write specialized computations where existing deep learning primitives do not provide enough control.
- Lower-level programming: Use Gluon when kernels need closer control over their implementation.
- Learning without a GPU: Run programming exercises through the Triton interpreter before moving to GPU execution.
The MIT-licensed project supports Linux. GPU execution requires NVIDIA hardware with Compute Capability 8.0 or higher, or AMD hardware with ROCm 6.2 or higher.
Logotyper
