Spaces:
Running
Hello! I have written my first kernel witu CuTeDSL and I would like authorization
Hello!!
I wrote my first kernel on softmax attention and have already this weekend made it far faster!
I am using CuTeDSL, and I am learning by doing, and I am currently an MSAI student
Here is the link to it: https://huggingface.co/monteirot/my_softmax_function
Hi Hugging Face team!
Bumping this request with a quick update.
I'm currently an MSAI student working on CUDA kernels using CuTeDSL.
This past weekend, I built and optimized a custom softmax attention kernel, significantly improving its speed.
I've already set up the initial repo here: https://huggingface.co/monteirot/my_softmax_function
Here is the code on GitHub: https://github.com/tiagomonteiro0715/my_softmax_function
I’d love to get kernel publishing access so I can share the new optimized implementation with the community!
Thanks for your time!