dslim

/

distilbert-NER

like 16

Your other models (bert-base / bert-large) have the nine token labels in the config listed as O, B/I-MISC, B/I-PER, B/I-ORG and B/I-LOC, but this distilbert model just has them as LABEL_0 to LABEL_8 (and from the look of the Inference API examples these are not in the same order as the bert models - LABEL_1 appears to match B-PER and LABEL_5 appears to be B-LOC).

Would you consider releasing an updated config with the correct B/I/O labels so the ner pipeline is able to aggregate tokens into chunks?