What is Multi-head attention?
Multi-head attention is an architectural concept in transformer-based language models. It is part of the vocabulary teams use when they compare models, design prompts or review an AI system.
Definition
Multi-head attention is an architectural concept in transformer-based language models. It is part of the vocabulary teams use when they compare models, design prompts or review an AI system. The term is used in architecture discussions, where precision matters: two teams can use the same word to mean slightly different things. In this glossary, multi-head attention is treated as a practical concept rather than a formal specification.
Why it matters
Architecture terms describe how a model represents tokens, moves information between positions and scales with parameters and data. They explain why two models with similar parameter counts can behave very differently.
Example
Suppose you are comparing Reka Core Nano with another model for a production task. Multi-head attention is one of the factors behind the numbers you see: Reka Core Nano lists a 16,384-token context window, costs $5.45 per million input tokens and scores 59.9 on GPQA Diamond. Understanding the term helps you decide which of those differences actually matter for your workload.
How to evaluate it
- Check whether multi-head attention is defined the same way in the documentation you are reading.
- Ask which measurement or trade-off the term is describing before comparing numbers.
- Separate vendor claims from independently observed behaviour on your own data.
- Write down the definition your team uses so reviews and evaluations stay consistent.
Related terms
Related guides
FAQ
What does Multi-head attention mean?
Multi-head attention is an architectural concept in transformer-based language models. It is part of the vocabulary teams use when they compare models, design prompts or review an AI system.
Why does Multi-head attention matter?
Architecture terms describe how a model represents tokens, moves information between positions and scales with parameters and data. They explain why two models with similar parameter counts can behave very differently.
How is Multi-head attention used in practice?
Suppose you are comparing Reka Core Nano with another model for a production task. Multi-head attention is one of the factors behind the numbers you see: Reka Core Nano lists a 16,384-token context window, costs $5.45 per million input tokens and scores 59.9 on GPQA Diamond. Understanding the term helps you decide which of those differences actually matter for your workload.