Launch your training function inside a notebook. Currently supports launching a training with TPUs on [Google Colab](https://colab.research.google.com/) and [Kaggle kernels](https://www.kaggle.com/code), as well as training on several GPUs (if the machine on which you are running your notebook has them).
An example can be found in this notebook.
Accelerator object should only be defined inside the training function. This is because the
initialization should be done inside the launcher only.