Deep learning

Current implementation of Go binding can not specify options.

GPUOptions struct is in internal package. And go generate doesn't work for protobuf directory. So we can't specify GPUOptions for NewSession.

Model description

I would like to add a new model:

Proposed in the paper: UNETR: Transformers for 3D Medical Image Segmentation

UNEt TRansformers (UNETR) utilize a transformer as the encoder to learn sequence representations of the input volume and effectively capture the global multi-scale information, while also following the successful "U-shaped"

🚀 The feature, motivation and pitch

The current implementation of Zero Redundancy optimizer has its own implementation of object broadcasting.

We should replace it with c10d [broadcast_object_list](https://pytorch.org/docs/stable/distributed.html#torch.distributed.broadcast_object_lis

What could be better?
依赖项的 requ…….txt 应该被更新，而且 pytorch 更要描述具体版本和安装方式

本人配环境一个下午，勉强配出来 Tag0.0.1 pytorch 1.9.0 OnlyCpu 的版本。

@babysor @oceanarium

This is a nice first issue:

Add types/docments to the Learner.get_preds function. This function is essential to any fastai user and has almost no documentation. Add types and text to the variables as we have in many places now.

What happened + What you expected to happen

When initializing a Ray Trainer, we provide a logdir argument, and the __init__ method of the Trainer stores it as a logdir class variable.

Then, when creating a Trainable with Trainer.to_tune_trainable(), it in-turn calls _create_tune_trainable(), which does not use self.logdir. So when tune_function is defined inside `_create_tu