Fix tok2vec-less textcat generation in website quickstart

[adrianeboyd]

Description

Something is going wrong with the expression conversion or evaluation on the javascript side, but separating the expression into smaller components seems to at least temporarily fix the problem.

Types of change

?

Checklist

• I confirm that I have the right to submit this contribution under the project's MIT license.
• I ran the tests, and all new and existing tests passed.
• My changes don't require a change to the documentation, or if they do, I've added all required information.

[almeidava93]

Hi!
I've got the same error here described, but with span categorisation. I didn't find yet a workaround. It doens't work, no matter if I use the pt_core_news_sm model or the pt_core_news_lg model. It seems to be related to tok2vec.

✘ Config validation error Bad value substitution: option 'width' in section 'components.spancat.model.tok2vec' contains an interpolation key 'components.tok2vec.model.encode.width' which is not a valid option name. Raw value: '${components.tok2vec.model.encode.width}'

Here is my code in the CLI:
python -m prodigy train models/spancat_rfe --base-model pt_core_news_lg --spancat test_3 --lang pt -V --config models\spancat_rfe.cfg

If I just remove tok2vec from the pipeline at the base-config.cfg file, I get another error:
✘ Config validation error Bad value substitution: option 'seed' in section 'training' contains an interpolation key 'system:seed' which is not a valid option name. Raw value: '${system:seed}'

I would appreciate some help with this. Thanks!

[adrianeboyd]

Hi @almeidava93 , an older/closed spacy PR isn't the best place for a new prodigy question. Since it's probably related to how prodigy is generating the spancat config internally, can you open a new thread in the prodigy support forum instead (https://support.prodi.gy)?