A ground-up technical course on encoder-style language models — the BERT family. Starting from how encoders differ from decoders, it works through bidirectional self-attention, the encoder block stack, masked language modeling, how representations become predictions, pretraining dynamics, and fine-tuning for downstream tasks. Ends with a practical landscape of the encoder family (RoBERTa, ALBERT, ELECTRA, DeBERTa, ModernBERT).
Audience: familiar with decoder-style LLMs at a high level; new to encoders. Prerequisites: linear algebra, multivariate calculus, basic probability. The LLM course is a natural companion.
Appendices
| A | A Field Guide to the Encoder Family |
| B | Encoder vs. Decoder |
| C | Acronyms |