4 ms·
Do they even have an optimized backward? It looks like optimizations like this aren't needed during training. Their V2 paper also suggests so.
by rfoo 2y ago
Do they even have an optimized backward? It looks like optimizations like this aren't needed during training. Their V2 paper also suggests so.