processor.tokenizer VS processor.feature_extractor,in [TrOCR] model

[Mohammed20201991]

I have a doubt about the init. trainer from @NielsRogge Transformers-Tutorials (TrOCR model)
Which one could be the correct value for passing to the tokenizer? processor.tokenizer VS processor.feature_extractor ,

from transformers import default_data_collator

# instantiate trainer
trainer = Seq2SeqTrainer(
    model=model,
    tokenizer=processor.feature_extractor,
    args=training_args,
    compute_metrics=compute_metrics,
    train_dataset=train_dataset,
    eval_dataset=eval_dataset,
    data_collator=default_data_collator,
)
trainer.train()

[Mohammed20201991]

processor.featuer_extractor

[CYaiche]

Ok but why ? The name of the parameter is tokenizer and you give the feature_extractor. This does not makes sens, does it ?

[UoFallujah]

@CYaiche Yea, you are right, that is why I am asking I think it was in old used version, processor.feature_extractor and used in the mentioned tutorial