chinakook commented on issue #18643:
URL:
https://github.com/apache/incubator-mxnet/issues/18643#issuecomment-662876105
A very big and complex graph with dynamic ops may be fail to delayed infer.
You can modularize them and then compose modules.
chinakook commented on issue #18643:
URL:
https://github.com/apache/incubator-mxnet/issues/18643#issuecomment-662868075
I also encounter this problem, delayed infer may be fail.
This is an automated message from the Apache
chinakook commented on issue #18643:
URL:
https://github.com/apache/incubator-mxnet/issues/18643#issuecomment-654293552
Actually, Loss is used explicitly, so it can be coded as Block, not
HybridBlock.
This is an automated
