Closed ghost closed 10 months ago
.
It is supported on Metal, but not on CUDA yet. I am mostly enable CUDA side of the SDP to see the performance differential and help to implement some other LLMs.
.