Hi,
I want to use the following HF model *[tomaarsen/span-marker-bert-base-acronyms]* to jump start NER on my dataset. This model works perfectly with spaCy's SpanMarker integration, but I can't seem to get it to work with Prodigy's hf.train.ner or hf.correct.ner. 1. Is this possible? 2. If so, what is the correct recipe? (Both Prodigy and SpanMarker reside in the same environment.)
Thanks
Welcome to the forum @Hellyeah 
!
SpanMarker models require span_marker python library for initialization, training and inference.
The hf.train.ner and hf.correct.ner use vanilla transformers library for model handling, so no, it won't work out of the box.
The good thing is that the entire plugin is open-source so you should be able to adapt it to use SpanMarker. I think this resource should be helpful to get it right quickly.
If your goal is to bootstrap your NER annotations and you already put together a spaCy pipeline with SpanMarker, another alternative workflow would be to use Prodigy model-as-annotator ner.model.annotate recipe.
If you need to adjust the recipe due to span_marker component having a different internal structure from the built-in spaCy components you can access the source code of the recipe in your Prodigy installation path. You can double check the Prodigy installation path by running prodigy stats command.