Adam operator takes too much time in CPU and lanch multiple kernels in CUDA #6985
Description
It could be slow. And we need a simple and unify way to write elemwise code.
{{ message }}
It could be slow. And we need a simple and unify way to write elemwise code.