Sasha Rush

Ah that's a great answer, thanks. Kind of suspected that was what was going on. Torch-Struct always assumes full materialization (except for CKY). Maybe we should have other version. CC...

Right... I know you could do it this way, but most of the time we materialize all the edges anyway for the potentials. However, for approaches like HMMs this is...

@teffland with https://github.com/harvardnlp/pytorch-struct/pull/103/commits things should now be fixed in torch 1.9. Can you update this PR to just change the base Entropy (and possibly cross-ent and KL) to use your...

Oops, will fix if I do a new version.

no I should just do a v2 this summer. lots of small fixes abound

I might be able to add this one if I find time

Can you let us know what is unclear about the visualizations? I think that would be useful for us.

Oh weird. I wonder if this is Gradio actually. I think it uses a little webui internally.

There is 😃

Thanks! I will take a look and update