Sean (Seok-Won) Yi

Also following.

I believe this is standard for preprocessing the BC5CDR data. "L2R" stands for "left to right" and vice versa. That means the if the triplet is "L2R" then the first...

Unfortunately there are a few places in the repository where the code conflicts with the paper. I'm assuming that when the authors say they "dropped" the query portion, what they...

+1 I'm also trying to install PyTorch XLA using Poetry but am finding it difficult due to `--find-links` not being available.

You should look at the original Transformer paper and other blog posts (e.g., The Illustrated Transformer is great) for some more information. The reason is because in self-attention we're performing...

I'm also getting the same assertion error problem. Does the model always have to return a dictionary? I'm following the examples provided in Label Studio and it seems like lists...

@chensun Sorry for the late reply. The versions are correct.

@rimolive Unfortunately I don't have access to the relevant resources anymore, so I don't think I'd be able to attempt that for the time being. It seems like a lot...

This probably isn't really an issue with the code itself. Check this Stack Overflow question: [UndefinedMetricWarning: F-score is ill-defined and being set to 0.0 in labels with no predicted samples](https://stackoverflow.com/questions/43162506/undefinedmetricwarning-f-score-is-ill-defined-and-being-set-to-0-0-in-labels-wi).

@seasonyao I don't recall seeing work that directly investigates the BERT > RoBERTa phenomenon, but I do recall that the authors of the [K-Adapter](https://aclanthology.org/2021.findings-acl.121/) model provide a possible explanation in...