Hi,
I am creating resnext and xception models on pytorch. Both models use grouped convolution which is supported in CUDNN 7. Does current pytorch support this? Thanks.
Hi,
I am creating resnext and xception models on pytorch. Both models use grouped convolution which is supported in CUDNN 7. Does current pytorch support this? Thanks.
If you build PyTorch from source and have cuDNN 7, you’ll get fast grouped convolution support in cuDNN. PyTorch v0.2.0 uses a slower for-loop over each group.