Hacker News new | ask | show | jobs
by thatguysaguy 810 days ago
Sounds like these guys didn't use custom kernels, but BitNet did.
1 comments
mobicham 809 days ago
That's correct. Only the dequantization is done on CUDA, the matmul is done with Pytorch. If they put their kernels open-source we could re-use them!
link