pretrain/PyTorch/configuration.py (2 lines):
	- line 4: # TODO better json handling
	- line 9: # TODO improve this implementation


pretrain/PyTorch/distributed_apex.py (2 lines):
	- line 1: # TODO: This is a copy of apex code from NVIDIA/APEX. Details to be added on the updates made here.
	- line 340: # TODO:  How do we want to handle multiple backward passes between


pretrain/PyTorch/train.py (1 line):
	- line 334: # TODO use logger


pretrain/PyTorch/azureml_adapter.py (1 line):
	- line 19: # TODO make this parameterizable