zh_core_web_lg static embedding come from where?

[rgib37190]

I want to know zh_core_web_lg static embedding come from where? Glove or fasttext? Thanks!!!

[polm]

I believe the current vectors are the same as in v2, and are trained on Wikipedia and OSCAR (Common Crawl) using fasttext.

https://v2.spacy.io/models/zh#zh_core_web_lg

This info got omitted when building the 3.0 models, but we'll be putting it back for 3.1.

[rgib37190]

Thanks for reply!!! But I see it is trained on OntoNotes. Which we use?

[polm]

The models (Parser, tagger, NER) are trained on OntoNotes. The vectors are trained separately and do not use OntoNotes.

[rgib37190]

I got it. Thanks for your reply!!!