3 CUDA GPUs are available; selecting 'cuda:2' (47622258688 bytes free).
Training...
  Epoch 0 / 199
    Time since training start: 0:00:29.244856
    Training data accumulated accuracy: 1.47 %
    Validation data accuracy: 4.41 %
    Closing learning rate: 0.001
    Accumulated training loss: 295.82687425613403
    Accumulated validation loss: 294.55209827423096
    Checkpoint? No