Just a quick note: we were delighted to see Kernl featured in an OpenAI Research Acceleration Team slide deck about Triton. The deck highlights Kernl as:
“Kernl - The first OSS inference engine written in Triton”
You can download the slides locally here: /files/slides/triton_intel.pdf (source: https://www.jokeren.tech/slides/triton_intel.pdf, OpenAI Research Acceleration Team, presenter: Keren Zhou).
It’s always motivating to see community work being recognized. Kernl started as a practical way to bring fused GPU kernels into everyday PyTorch workflows using the OpenAI Triton language, aiming for strong speedups without leaving the comfort of Python. If you’re curious about the background and benchmarks, our earlier post gives the full story: Up to 12X faster GPU inference on Bert, T5 and other transformers with OpenAI Triton kernels.
Many thanks to the Triton and OpenAI teams for building a tool that makes this kind of work accessible-and to everyone in the community who keeps pushing the ecosystem forward.