Dropout types for Bart Model

Hi,

I am a beginner and currently fine-tuning a BART Model (ForConditionalGeneration) for text summarization. I wanted to experiment with dropout however noticed that the model has different times of dropouts kin its configuration, namely attention_dropout

  • activation_dropout
  • classif_dropout
  • classifier_dropout
  • dropout

I could not find any documentations on the different types and which best to use and hope I am correct in this forum. I only found that the default values for all of them except the basic dropout is 0.0. And was wondering if this has a special reason for the application? Can anybody give me some guidance here?

Thank you in advance