← Back to Glossary
Transformer
Phase 02 · LLM Fundamentalscore-concept

The foundational neural network architecture behind modern LLMs that processes all words in parallel using self-attention.

Why it exists

Older models (RNNs) processed text word-by-word sequentially. Transformers process entire text sequences in parallel, allowing massive scaling on GPUs.

How it works

Uses stacked encoder/decoder blocks with multi-head self-attention to link related words across long sequences and feed-forward layers to refine representations.

Where you'll use this