MVP example for multiple GPU training

[pdogra89]

-Additional refactoring iteration to unify single-gpu/multiple gpu training?

[Nic-Ma]

Hi @madil90 ,

As you are expert on Deep Learning performance tuning and also done many investigations on PyTorch, could you please help to develop this task?
The minimal requirements will be:

1. Wenqi extracts pure python example from notebook example(already committed PR).
2. You can submit a PR to add Multi-GPU logics for training and evaluation(still in development) examples.
3. BTW, please note that: don't save the parallel model to files, be careful.

Please feel free to contact if you have any questions.
Thanks in advance.

[pdogra89]

Eric will create a work in progress PR - Eric and Adil to align - don't want to independent pull requests.

[wyli]

resolved via #49 #51