triton-lang/triton

Health v2 · maintenance only

Adoption not part of the score

20386 stars · 3302 forks observed · 2026-10-17

What it is AI-extracted, prompt v1, taxonomy v1, 2026-10-19, confidence 0.97

Triton is an open‑source programming language and compiler for writing high‑performance custom deep‑learning primitives. It targets GPUs and provides a Python‑friendly API to generate optimized kernels without writing CUDA code.

Use cases

• write custom GPU kernels for deep learning models
• accelerate matrix‑multiplication or convolution operations with a DSL
• prototype new neural network layers faster than using raw CUDA
• optimize performance‑critical sections of PyTorch or TensorFlow pipelines
• experiment with novel tiling strategies for tensor computations

When to choose

• you need fine‑grained control over GPU kernel performance while staying in Python
• existing frameworks lack a specific operation and you want to implement it yourself
• you prefer a higher‑level DSL over writing low‑level CUDA or PTX
• your project targets NVIDIA GPUs and you want to leverage MLIR‑based optimizations

When to avoid

• you require cross‑vendor GPU support (e.g., AMD or Intel GPUs) as Triton currently targets NVIDIA only
• your workload runs on CPUs only or on non‑GPU hardware
• you need a fully managed high‑level API without writing any custom kernels
• you are constrained to environments where installing Python wheels is not possible

Facets

library · maturity active

compiler deep-learning gpu-computing artificial-intelligence deep-learning machine-learning windows python dsl custom-kernels cuda-alternative high-performance tensor-operations mlir linux macos gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="triton-lang/triton")