3 ms·
Seconded. I’m guessing you could create an implementation that is able to do that and then write optimised triton/cuda kernels to accelerate them but need to in
by diwank 2y ago
Seconded. I’m guessing you could create an implementation that is able to do that and then write optimised triton/cuda kernels to accelerate them but need to investigate further