triton-materials

This repo is a scratch pad for exploring OpenAI's Triton

Notes

Remote development in a Docker container on Lambda Cloud with VS Code:

First, you need to add your user to the docker group:

sudo usermod -aG docker $USER

Then, close and reopent the VS code connection.

Now you'll be able to initialize a remote session in the container. But, first:

sudo docker pull joehoover/kernl:latest
git clone https://github.com/joehoover/triton-materials.git
sudo docker run --rm -it --gpus all -v $(pwd):/triton-materials joehoover/kernl:latest

Installation

Lambda Cloud instance had a stale version of Triton. This (issue)[triton-lang/triton#625 ] suggested updating with:

pip install --pre -U triton

Useful Links

CUDA refresher series

Triton-lang (introduction)[https://triton-lang.org/master/programming-guide/chapter-1/introduction.html]

https://analyticssteps.com/blogs/openai-triton-programming-language-neural-networks