Single model multi GPU training

[BarKetPlace]

The discriminative step is now limited to the capacity of one GPU
We must be able to perform the training of all single models on several GPUs.

[BarKetPlace]

In fact the discriminative training steps requires to much memory to run on a single gpu. So this becomes a problem of it s own