No GPU usage in model training

I have installed TitanXP Nvidia drivers on Ubuntu 16.04.4 LTS platform following all the commands at:

I have pulled the latest tensorflow NGC container ( 18.06-py3)
The nvidia-smi command correctly reports the card.
However, when I run a Keras model, it seems that it is run only in cpu, as the nvidia-smi command shows no gpu utilization!!
How can this be happening? Any hints??

The requirements.txt from the DL model I try to use, installed programs (probably Keras) which broke the proper directions for using the gpu…

Problem solved.