Spaces:
Running
Running
Kernel publishing access request: Nousphera VPTQ CUDA kernels
#38
by vtsouval - opened
Hi, I’m Vasilis. We’re developing reusable VPTQ inference kernels for low-bit language models, including packed codebook lookup and grouped GEMM targeting RTX 5090 / Blackwell and Ampere GPUs. We’d like to distribute the compiled kernels through Hugging Face so applications can install them using the kernels library.
Could you enable kernel publishing for Nousphera? We’ve submitted the organization access request. The source is currently private; kernel-builder integration and RTX 5090 correctness tests are in place. Thank you!
Hi! Thanks for reaching out! We usually ask for some public some source repo to do some sanity checks before enabling the kernel repo type.