We have released Neural Network Libraries 1.0.20!
ShuffleNet and ResNext are now available as pre-trained models!
Furthermore, pre-trained models have become compatible in dynamic mode, as well as directly callable!
nn.set_auto_forward(auto=True) # switch to dynamic mode from nnabla.models.imagenet import ResNet34 model = ResNet34() batch_size = 1 x = nn.Variable((batch_size, ) + model.input_shape) x.d = ... # set some value y = model(x) # works as expected
- Add CUDA stream handler class in Python (as an unsupported feature so far)