Triton
OpenAI's open-source language for writing GPU kernels; makes it easier to write custom CUDA code without low-level details.
Think of it like
High-level GPU programming; closer to Python, farther from CUDA.
Example
Write a fused attention kernel in Triton; compiles to optimized CUDA.
How it actually works
Triton abstracts memory layouts and synchronization; code is faster to write and still efficient. Enables rapid experimentation with new operations.
For product teams
Reduces time to optimize kernels; speeds up research.
For engineers
Triton kernels are fast and maintainable; becoming standard for LLM-specific optimizations.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome