Inference with Incremental Model

[apapoudakis]

Hello,

Congratulations for your work which is super interesting!

Can I have instructions (and probably the checkpoints) on how to run the incremental approach on longer texts?

Thank you in advance for your help!
Argyris

[g185]

Hi @apapoudakis,
Sorry for the late reply!

I have been postponing the upload of those models for a while since I have been working on some other projects, which specifically address the lack of long document models.
I will work on sharing the incremental model weights in the upcoming days.

If you need it asap, I suggest you to train a system with the incremental formulation. It is enough to change the model configuration in the conf/root.yaml from mes/deberta-large to incr/deberta-large, and train a new model from scratch following the instructions in the README.

PS. I have just read your recent paper BookWorm and congrats for the very interesting work!
If you are interested in the book-length setting, we are working on a similar direction for Coreference Resolution which we will release soon.