Load (PyTorch|Tensorflow) model, add or drop layers, and finetune

[kgoderis]

I am looking for an example that

1. Loads an existing model, for example a PyTorch from HuggingFace
2. Drops some layers from that model
3. Add some new layers to that model
4. Trains that model

Can an existing model be embedded in a new model?

[zachgk]

Take a look at our transfer learning examples such as the ResNet Cifar10 example. However, support for removing the last block is only available in MXNet right now