Hugging Face's Auto classes automatically detect the right architecture and load the correct model class — so your code works across any model.

Key Auto Classes

from transformers import AutoTokenizer, AutoModel, AutoModelForSequenceClassification

# Works with any model — bert, roberta, distilbert, etc.
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
model = AutoModelForSequenceClassification.from_pretrained("bert-base-uncased")

Common AutoModel Types

• AutoModelForSequenceClassification — text classification
• AutoModelForTokenClassification — NER, POS tagging
• AutoModelForQuestionAnswering — extractive QA
• AutoModelForSeq2SeqLM — translation, summarization
• AutoModelForCausalLM — text generation

Reference:

Auto classes documentation

https://huggingface.co/docs/transformers/model_doc/auto