Text Classification Without using Auto Model For Sequence Classification

Hi every, current I’m working with a project to classify texts to analyze mental health and I notice that most of the models can use with AutoModelForSequenceClassification.

However, I have went to huggingface public models such as Mental-Flan-T5, Mental-Alpaca, or even Mental-Bert, I saw they used AutoModelForSeq2seq, AutoModelForCausalLM, or AutoModelForMaskedLM.

What I’m confusing here is that how can they classify texts without the classification head (and why don’t they do that), and also how to evaluate the result of the model if it generate the text instead.

Thank you very much.