John Bauer

Idea - rebuild glove vectors

1

Goal: build new glove vectors with current vocab - new people you haven't heard of should now have word vectors - match tokenization of the glove vectors with the upcoming...

It appears 5.10 is available on maven, but README.md still has xml for connecting to 5.8 on maven

No convenient way to apply class weights for a segmentation model

10

- Have I written custom code (as opposed to using example directory): The custom code is to build a segmentation model: U-Net model, go from (batch, x, y, channels) ->...

There's a wandb integration available for java. It might be nice to add, but only if it can silently ignore wandb when the python wandb isn't installed.

Add a script to convert from AWS annotator reports to a report on how much work each annotator did

Transliterate kazakh to a latin alphabet

Automatically extract text & sentiment?

2

Is there a convenient command line tool or a sequence of mmax calls which will extract the spans and the sentiment polarity? The best I figured out so far was...

There are two actions done in `save_pretrained()`: https://github.com/huggingface/peft/blob/2822398fbe896f25d4dac5e468624dc5fd65a51b/src/peft/peft_model.py#L98 save_pretrained builds an `output_state_dict` of just the items that will be saved, and then it writes those items to a directory. It...

How to use PEFT to wrap an encoder?

1

For the use case of training downstream models which aren't exactly token classification or sequence classification, I have been using Bert, Roberta, etc models as encoders to get context sensitive...

Only include `labels` as a parameter if it's specified in the functio…

2

Only include `labels` as a parameter if it's specified in the function call. This way, encoders as opposed to classifier models can also function. Currently only implemented for the TOKEN_CLS...