Decoder
The half of a model that generates output one token at a time, each based on what came before.
Think of it like
A writer producing a sentence word by word, each choice shaped by everything already written.
Example
GPT and Claude are decoder-only models: they read your prompt and generate the continuation token by token.
How it actually works
A decoder produces tokens sequentially, and to keep generation honest it uses causal (masked) attention so each position can see only earlier tokens, never future ones. Most of today’s large language models are decoder-only — a single stack that both reads the prompt and writes the answer. It’s the architecture built for generation.
For product teams
The shape behind chat and text-generation models — read a prompt, write a continuation.
For engineers
A stack that generates tokens autoregressively using causally-masked self-attention.
Related
- Encoder — The counterpart to an encoder.
- Autoregressive — Generates autoregressively.
- Causal Attention — Uses causal attention.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome