TensorFlow prints log information twice

Question

I'm running TensorFlow official model from https://github.com/tensorflow/models/tree/master/official

When I run the official resnet model, each tensorflow INFO message is printed twice as displayed below:

INFO:tensorflow:cross_entropy = 2.053623, learning_rate = 0.025, train_accuracy = 0.19732143 (0.102 sec)
I0927 11:58:25.770764 140597122463552 basic_session_run_hooks.py:247] cross_entropy = 2.053623, learning_rate = 0.025, train_accuracy = 0.19732143 (0.102 sec)
INFO:tensorflow:cross_entropy = 1.7440678, learning_rate = 0.025, train_accuracy = 0.19938381 (0.107 sec)
I0927 11:58:25.878196 140597122463552 basic_session_run_hooks.py:247] cross_entropy = 1.7440678, learning_rate = 0.025, train_accuracy = 0.19938381 (0.107 sec)                                                                                   
INFO:tensorflow:cross_entropy = 2.116714, learning_rate = 0.025, train_accuracy = 0.1983507 (0.102 sec)
I0927 11:58:25.980251 140597122463552 basic_session_run_hooks.py:247] cross_entropy = 2.116714, learning_rate = 0.025, train_accuracy = 0.1983507 (0.102 sec)

My current log level is

export TF_CPP_MIN_LOG_LEVEL=2

Can someone help with the issue? Thanks!

It's in your training code. When you get tensorflow logger, set this property to false. — zihaozhihao, Sep 27 '19 at 17:45
Thanks, however I couldn't manage to find this parameter. I'm using TF-1.13. Could you please share some doc for this? — Jon.W, Sep 27 '19 at 18:04
Could you check this post, looks like they have the same issue as yours. https://stackoverflow.com/questions/33662648/tensorflow-causes-logging-messages-to-double — zihaozhihao, Sep 27 '19 at 18:08

TensorFlow prints log information twice

0 Answers0