may ohta

I thought $\mathcal{O}(\max(|src|^2, |trg|^2, |src|*|trg|))$, which corresponds to src-src attention, trg-trg attention, and src-trg attention, respectively. So, essentially the same as the first one: $\mathcal{O}(\max(|src|, |trg|)^2)$. (Ignored the factor of...

Hi @tongye98 Thank you for asking! The short answer is no, simply because I cannot make it work. I tried to integrate `DistributedDataParallel`, but for some reason I keep getting...

Reminder for myself: - https://pytorch.org/docs/stable/notes/ddp.html - https://pytorch.org/tutorials/intermediate/ddp_tutorial.html - https://pytorch.org/tutorials/intermediate/dist_tuto.html

Hi @wjbyrne, Thank you for reporting an issue. [joey_demo.ipynb](https://github.com/joeynmt/joeynmt/blob/main/notebooks/joey_demo.ipynb) was written for joeynmt v1.x, not compatible with joeynmt v2. Could you please check your joeynmt version? Sorry, we should have...

Oh, I see. Some bug fixes in joeynmt are not reflected in PyPI server yet, I guess. We will check it. Thank you for letting us know!

TODO: adapt README, tutorials and documentation.

@juliakreutzer could you wait for a few days, maybe? I'm preparing an update to pytorch v1.12, since the official google colab also has the default torch version 1.12, now. Could...

but at the same time I'm a bit afraid, because it seems torch 1.12 is kind of a big jump from 1.11. More concretely, cuda116 (for torch 1.12) has a...

I merged the change in the demo notebook https://github.com/joeynmt/joeynmt/pull/197. In the notebook, we install joey from the repo with the version number specified, so that we can avoid the version...

Awesome! Should we maybe save the value of `self.stats.validation_steps_since_last_improvement` in the state_dict to make it resumable, when someone want to continue training from a saved ckpt? just like `self.stats.best_ckpt_score`?