William Falcon

nice! yeah, when you enable tensorboard and checkpoints the training slows down a bit. we actually have tests to ensure we don’t incur a meaningful overhead over a vanilla pytorch...

next you should convert the datasets to datamodules. this means you can 100% decouple the model from the data :)

ah i see the confusion! ### what is stage in setup? .fit() calls setup (for train, val data) and .test() also calls setup (for test only). Our data calls are...

> Alright calling it here for today, I'm tired and still have some actual work to do. I'm pretty sure I don't understand how `pl.LightningDataModule` API is supposed to be...

ok, i understand the confusion! #### doc updates I updated the docs to show results as an optional extension! Also split the docs into optional vs required. [Doc updates here](https://pytorch-lightning.readthedocs.io/en/latest/new-project.html)...

yup, happy to add it... but by adding lightning you don’t need the train.py file because you can copy paste the code from the jupyter into a script and run...

awesome! btw, the code internals are currently going through a refactor we started a few weeks ago in preparation of 1.0. Our goal is to get the internal to read...

Why not just use [PyTorch Lightning](https://pytorch-lightning.readthedocs.io/en/latest/new-project.html) for your trainer and code structure? We already support GPUs, multi-GPU, multi-node, 16-bit (apex and native), and about 40+ other things. Lightning is already...

is “show” the best command word here? @adam-lightning. how will this generalize as the CLI evolves?

that makes sense. i'm just thinking about what else "show" will apply to other than logs