FP8 and FP4
#5
by
whatever1983
- opened
Just a suggestion for @Nvidia:
When you release awesome models like this, can you also release FP8 and FP4 versions? Of course FP8 is to run on the H100 and FP4 is to run on GB10 and GB200/GB300s.
Conversion to FP8 and FP4 might take a while and Nvidia should do us the favor. NIMs should also denote the FP8/FP4 difference.