New Features:
- Add support for Torch recurrent layers (RNN, RNNCell, LSTM, LSTMCell, GRU, GRUCell)
- Add support for device errors caused by nonlinear I-V response
- Update example ResNet50 scripts in /applications to use Torchvision-native dataset loaders.
Note: These new scripts may change the order in which images are applied. For experiments using a subset of the ImageNet dataset this reordering may change results. 3.1.x examples scripts are fully compatible with 3.2 and should be used if exact reproducibility for subsets of the ImageNet dataset is required. We expect future example scripts to follow the new Torchvision-native structure and encourage users to switch to these scripts."