Problem description

In tensorflow GPU 1.14.0 environment, using multiple GPU training will report error
change to tensorflow GPU 1.13.1 to solve this problem