cardiffnlp

/

twitter-xlm-roberta-base-sentiment

like 187

Follow

Cardiff NLP 54

Hi,

loading the model with protobuf > 3.20.0 fails.
As suggested in the stacktrace one can set the env variable
PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python
to prevent the error.

An alternative is to load the model once with the env setting and the new protobuf version and store it again (model.save_pretrained()). The newly serialized model will than load in a new protobuf version without the env variable.