onrup

Model architecture

Attention

What is attention?

Attention is the mechanism that lets a model weigh the relevance of every earlier token when processing the current one. Its cost grows quadratically with sequence length, which is why long context is expensive.

Each token computes a query, and every token produces a key and a value. Comparing queries against keys is what makes cost scale with the square of the sequence.

Grouped-query attention, which shares keys and values across several query heads, is the standard mitigation in modern models and is a large part of why long contexts became practical.

Related terms

All terms in the glossary →

Start with the free tier

A magic link creates your account, your tenant and your first API key. No card until you ask for compute.