Transformers Explained
A step-by-step blog and video series explaining the Transformer architecture, covering tokenization, embeddings, every attention variant, feed-forward layers, and how the encoder and decoder fit together.
Transformers Explained walks through the Transformer architecture piece by piece rather than treating it as a black box. Part 1 covers the complete architecture: why the Transformer was needed, tokenization, embedding, and positional encoding, the attention mechanism and multi-head attention, feed-forward networks, residual connections and layer normalization, how the encoder and decoder work together, and the three architectural variants.
Part 2 goes deeper into attention specifically, starting from the intuition behind self-attention and building up to the exact query, key, and value math with a worked numeric example, then continuing into multi-head self-attention and where it is used in modern models such as BERT and GPT. The series is being written and expanded over time by its author as new posts and videos are produced.
It is maintained by the founder of an online learning platform as ongoing educational content rather than a software library.