[E896] on training existing model (NER)

Hi, i'm running the train recipe, but getting this error:

RuntimeError: [E896] There was an error using the static vectors. Ensure that the vectors of the vocab are properly initialized, or set 'include_static_vectors' to False.

The command im using is:

python -m prodigy train ./nercorrectoutputdir --ner prs_correction_dataset --base-model sv_core_news_lg

Been trying to search any information about this, but come up short.
Any ideas how to fix this issue?

hi @aheydeck!

Thanks for your question and welcome to the Prodigy community :wave:

Just curious, what if you run:

python -m prodigy spacy-to-data ./nercorrectoutputdir --ner --ner prs_correction_dataset --base-model sv_core_news_lg 
python -m spacy train ./nercorrectoutputdir/config.cfg --paths.train ./nercorrectoutputdir/train.spacy --paths.dev ./nercorrectoutputdir/dev.spacy

prodigy train is just a wrapper for spacy train. data-to-spacy will break up a hidden step in prodigy train where it creates the training/config file and instead do the first step by outputting your data into a train/eval binary files (that's the train.spacy and dev.spacy) and create a default spaCy config file. You'd then run spacy train on the output of data-to-spacy (see the docs for more details).

I'm wondering if there's some issue with the config file. Your error message suggests a possible resolution where you can modify the config file by either initializing your vectors (e.g., add this to your config file vectors = "sv_core_news_lg") or include_static_vectors to False. So by using data-to-spacy, can you try either of these suggestions to modify your config.cfg file?

Does this help?