That’s definitely strange. Could you provide the commands, version and data file so I can replicate?
@honnibal could you share your email id? I am not very sure if i can share the annotated data in a public forum.
Ok, The same training file runs in Ubuntu. It breaks in mac.
In ubuntu, it runs, creates a model and saves it, but it does not tag a new sentence. This training data has about 1000 real time examples. Now, i don’t understand why is the model accuracy very low. (i got to see the accuracy by running spacy through prodigy ner-batch-train)
My email is firstname.lastname@example.org