The given usage code returns an error.

#2
by andres-pb - opened

Version: 4.25.1

I had a similar issue. It runs on version 4.2.0

Arabic Language Technologies, Qatar Computing Research Institute org

Thanks a lot for reporting this and @ElMad for the temporary work around. I tried poking around a bit with specific tokenizers and all of them seem to be giving the same error; I'll report this on transformers repo and see if they are able to state why this is happening/how we can update our code.

Will keep this issue open for now and update it when I have some news.

Arabic Language Technologies, Qatar Computing Research Institute org

The transformers issue is here: https://github.com/huggingface/transformers/issues/21240

There is a quick temporary fix posted there by someone else that works with the latest transformers version, so check it out while the underlying issue gets resolved.

Arabic Language Technologies, Qatar Computing Research Institute org
edited Jan 25, 2023

Its fixed @andres-pb ! The pipeline works best when the model/tokenizer are passed in as explicit keyword arguments (model=model, tokenizer=tokenizer). I've updated the usage code and tested with both the latest transformers and an older version (~4.1).

Feel free to comment/open another issue if you face any problems!

fdalvi changed discussion status to closed

Sign up or log in to comment