Thanks for the great repo!
Can you point me to what needs to be done to enable fp16 compatibility? I suppose it will need some rewriting in the CUDA kernels. Is it just changing the types in the function declarations or more work that needs to go into it?
Thanks for the great repo!
Can you point me to what needs to be done to enable fp16 compatibility? I suppose it will need some rewriting in the CUDA kernels. Is it just changing the types in the function declarations or more work that needs to go into it?