talschuster / cats Goto Github PK

Hello,

I'm interested in using your model as a baseline for my current project. However, I'm having difficulty setting it up due to the lack of clear instructions on training and evaluation using the provided code.

Would you be able to provide a set of bash commands that you use to train and evaluate your model on a specific dataset (e.g., IMDB)? This would help me greatly in replicating your work.

Thank you in advance!

Hi, and thank you for your great work!

I was wondering if the early exit techniques introduced in the paper can be extended to be used with language modeling, or do they only apply to classification tasks? I think the only difference is that (1) language modeling has a rather large answer space at tens of thousands of vocabularies, and that (2) language models usually output a probability distribution to be sampled. Maybe it is because the conservative predictions are not strong enough when facing such a large number of possible sampling outcomes?

I see that you have a later work (CALM) addressing the case on language models by enforcing the early-exit objective during training, but I think the approaches used in CATs are more desirable because it is distribution-free and model-agnostic.

Thank you for your time!