Bug description:
Hi I'm following the "models as annotators" tutorial, trying to use spacy-medium to label 4 entities. However the argument --labels seem to be wrong, and I cannot get --label to work as expected either.
Reproduction steps: How can we recreate the bug?
Firstly I tried run command exactly like how tutorial states, by passing --labels ORG,PERSON,LOC,PRODUCT
I then tried to pass by --label ORG,PERSON,LOC,PRODUCT. Note it's label rather than labels. It runs, but does not process any annotation, as you notice the progress states is 0%, rather than 100% like previous step.
There's definitely a mistake in the video tutorial around minute 9. I'm guessing it's because it was recorded with a version of the recipe that wasn't 100% final.
Sorry about that!
The right argument name is --label. Btw. you can always run the recipe with just the -h flag to get CLI help on the arguments.
The reason why none of the examples gets annotated on your second pass is that the ner.model-annotate recipe filters out the examples that were already annotated by the model as defined via model-alias argument.
In other words, since in your second command (with the subset of labels) you use the same target dataset and the same model-alias - all examples are filtered out because they are considered to be already annotated by en_core_web_md. So for your experiment to work, you should change the target dataset name (or model alias name) to get the recipe to annotate all examples again.
Bug description:
I'm following the "models as annotators" tutorial, trying to use spacy-medium to label 4 entities. However the argument 
Reproduction steps:
Environment variables: