- Support for SqueezeNet architectures
- Fix incompatibility with PyTorch 1.7 (#83)
- Allow user to specify only some device ids for training through the
dp_device_ids
argument totrain.train_model
- Update requirements.txt
dp_device_ids
argument to train.train_model