I am new to HF and I am I am still familiarising with classes config files etc
While playing with an EncoderDecoderModel I had this doubt:
robertashare = EncoderDecoderModel.from_encoder_decoder_pretrained(encoder_pretrained_model_name_or_path="roberta-base",
decoder_pretrained_model_name_or_path="roberta-base",
tie_encoder_decoder=True)
print(robertashare.config.encoder.architectures)
print(type(list(robertashare.children())[0]))
Here the first print logs “[‘RobertaForMaskedLM’]” and the second “<class ‘transformers.models.roberta.modeling_roberta.RobertaModel’>”
Why is it? I was expecting these two to match.
The same mismatch arises for the decoder:
print(robertashare.config.decoder.architectures)
print(type(list(robertashare.children())[1]))
The first print logs “[‘RobertaForMaskedLM’]” and the second class “<class ‘transformers.models.roberta.modeling_roberta.RobertaForCausalLM’>”
Anyone could clarify?
Thanks in advance