Kyle Gorman

> > If the user requests a non-shared embedding matrix, keep the source and target embeddings separate by enclosing the source symbols {like} {this} (cf. what we do with the...

I just noticed this is still open. Ha. Closed in #140, even if we're still working the kinks out.

@Adamits I have assigned this to you since you wanna take a stab. There is a new interface for CLI stuff (probably an imporvement over the somewhat ad hoc one...

I have read these docs and have the basic lay of the land now. The migration should go in two steps: 1. First, we should implement subclasses of `LightningDataModule` (this...

> Sounds good, I think I understand. Though I feel a bit annoyed that we have to use `LightningCli` instead of standard python libraries... I will try to prioritize it...

@Adamits believes it is possible to migrate to Torch 2 while staying on the 1.13 branch of PTL. The ability to complete this migration in two steps (one to Torch...

Draft of the migration to PyTorch 2 in #173. Testing now---focusing on the transformer since we can use the causal mask flag there, hopefully.

I have been using LightningCLI in another project and I have to say, it's a total vibe shift and it's way better than having big Bash scripts (without any nice...

Another feature of this migration is that at least some if not all the models will work on M1/M2 Macs using their MPS accelerators.

My only thought is I am most excited about variant 2. I thought the A&G thing was outmoded also, but it's harmless if you want to do it later.