BERT reads (encoder only), GPT writes (decoder only); pick by task.
- BERTEncoder that sees the whole sentence both ways.
- GPTDecoder that predicts the next token.
- Fine-tune or promptBERT is fine-tuned per task; GPT is mostly prompted.
- CostBERT is small and fast; GPT-class models are big.
Explain 1/5
Source: Hugging Face: BERT