Hello,
I'm Niels and work as part of the open-source team at Hugging Face. I discovered your excellent work on CUDA-L2 through Hugging Face's daily papers as yours got featured: https://huggingface.co/papers/2512.02551.
The paper page lets people discuss about your paper and lets them find artifacts about it (your optimized kernels for instance),
you can also claim the paper as yours which will show up on your public profile at HF, add Github and project page URLs.
Your CUDA-L2 optimized HGEMM kernels for A100 GPUs offer significant performance improvements for matrix multiplication, a fundamental operation in many AI workloads, especially LLMs. We believe that hosting these optimized kernels on https://huggingface.co/models could greatly enhance their visibility and discoverability within the broader AI community. This would allow users to easily find, download, and integrate your work. Even though these are low-level computational artifacts, their impact on the efficiency of ML systems makes them highly relevant for the Hub.
If you're down, leaving a guide here. For custom CUDA kernels, people can use hf_hub_download to fetch the compiled binaries or source code directly from the Hub.
After uploaded, we can also link the kernels to the paper page (read here) so people can discover your work.
You can also build a demo for your work on Spaces, we can provide you a ZeroGPU grant, which gives you A100 GPUs for free.
Let me know if you're interested/need any guidance :)
Kind regards,
Niels
ML Engineer @ HF 🤗
Hello,
I'm Niels and work as part of the open-source team at Hugging Face. I discovered your excellent work on CUDA-L2 through Hugging Face's daily papers as yours got featured: https://huggingface.co/papers/2512.02551.
The paper page lets people discuss about your paper and lets them find artifacts about it (your optimized kernels for instance),
you can also claim the paper as yours which will show up on your public profile at HF, add Github and project page URLs.
Your CUDA-L2 optimized HGEMM kernels for A100 GPUs offer significant performance improvements for matrix multiplication, a fundamental operation in many AI workloads, especially LLMs. We believe that hosting these optimized kernels on https://huggingface.co/models could greatly enhance their visibility and discoverability within the broader AI community. This would allow users to easily find, download, and integrate your work. Even though these are low-level computational artifacts, their impact on the efficiency of ML systems makes them highly relevant for the Hub.
If you're down, leaving a guide here. For custom CUDA kernels, people can use hf_hub_download to fetch the compiled binaries or source code directly from the Hub.
After uploaded, we can also link the kernels to the paper page (read here) so people can discover your work.
You can also build a demo for your work on Spaces, we can provide you a ZeroGPU grant, which gives you A100 GPUs for free.
Let me know if you're interested/need any guidance :)
Kind regards,
Niels
ML Engineer @ HF 🤗