How is the "Auto Model For Sequence Classification" architecture?

How is the architecture of AutoModelForSequenceClassification? I suppose it’s some pre-trained transformer with some dense layer for classification, however where could I see the forward details of this model?

1 Like

The auto classes are just abstractions that work for every architecture. You can see the actual forward passes in each modeling files. For instance, if you are using a BERT checkpoint, you will get a BertForSequenceClassification model, which forward pass is defined in `transformers.models.bert.modeling_bert

1 Like

Thank you @sgugger.