Not working with Bohek 3.0.0 about bulk HOT 7 CLOSED

Yes, I'm following the example you left in the readme for bulk, though I swapped from the 'all-MiniLM-L6-v2' model to the 'distiluse-base-multilingual-cased-v2' model because the tweets are mostly in Persian with some English and I've had success with that model in the past for multilingual data. There are about 190,000 tweets in total all about the protests and unhappiness with the government in Iran. I've gotten some really interesting results already, even though I only started with the first 10,000 tweets, so I appreciate the work you've put into bulk.

from bulk.

Ah, good call. Yeah that came out a few days ago and it seems that it broke something.

Will have a look, thanks for reporting!

from bulk.

I've set up a max version for now. That way, it should not break while I figure out what to do about v3 of Bokeh.

Can you check if v0.1.2 works out of the box now?

from bulk.

Yes, version 0.1.2 works out of the box now. Thanks for the updates!

from bulk.

Grand. Thanks for reporting!

from bulk.

Also, I'm a bit curious. What embeddings are you using for Twitter? Stuff from embetter?

from bulk.

Happy to hear it.

At some point, I might also recommend taking a step back from bulk and using a trained ML model and a proper labeling tool to proceed. Bulk is meant to help you get started, but the labels aren't 100% accurate, and I'd assume that a fine-tuned model might be of more help later. Just a gut feeling, though.

from bulk.