Details
-
New Feature
-
Status: To Do
-
Major
-
Resolution: Unresolved
-
None
-
None
Description
Since fused RNN operator for GPU is already implemented with cudnn, I think we should also provide the same functionality at CPU side. So user models which call sym.RNN can run both on GPU and CPU. This will also improve the performance of RNN related models a lot on CPU.
