mkolod commented on issue #11380: Add ability to query cuDNN BatchNorm min. epsilon. Allow ONNX importer to use cuDNN BN if chosen eps >= cuDNN min. eps. URL: https://github.com/apache/incubator-mxnet/pull/11380#issuecomment-401080568 @szha Makes sense. In that case, I'd propose the following. How about I remove the Python function to query that cuDNN constant and inline that call in the ONNX importer, so it's not exposed as an API if it's indeed a one-off? That said, even in that case, I'd need the actual C function because that constant is not a static const float that can be accessed, and it's not a cuDNN query function either. It's a macro, so it needs the C query function. The use in Python could be inlined though. I hope that we agree on the issue that it would be useful not to disable cuDNN BN for ONNX imports if the epsilon is big enough that it is equal to, or exceeds, the cuDNN minimum. cuDNN BN is faster for both fp32 and fp16.
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
