why ''last_layer_features = xlmr.extract_features(zh_tokens)'' just jump to the function of ''fairseq.models.roberta.hub_interface.RobertaHubInterface.extract_features'' ?
what the difference between ''fairseq.models.roberta.model.RobertaEncoder.extract_features'' and ''fairseq.models.roberta.hub_interface.RobertaHubInterface.extract_features'' ?
lematt1991
commented
3 years ago
Yep, that's what the classification heads do
Originally posted by @lematt1991 in https://github.com/pytorch/fairseq/issues/2302#issuecomment-654237044
why ''last_layer_features = xlmr.extract_features(zh_tokens)'' just jump to the function of ''fairseq.models.roberta.hub_interface.RobertaHubInterface.extract_features'' ? what the difference between ''fairseq.models.roberta.model.RobertaEncoder.extract_features'' and ''fairseq.models.roberta.hub_interface.RobertaHubInterface.extract_features'' ?