karpathy / llm.c

LLM training in simple, raw C/CUDA

Repository from Github https://github.comkarpathy/llm.cRepository from Github https://github.comkarpathy/llm.c

About pull request of custom kernel implementation

KarhouTam opened this issue · comments

Hey. First of all, I really appreciate this repo, it helps me learn cuda programming a lot.

I tried to implement a softmax forward kernel myself and mixed some optimization skills that I learned and found that it actually works well on my GPU (3070 laptop).

Does this repo accept PR for custom implementation of existing kernel?