Models
Attention head
An attention head is one parallel attention mechanism inside a transformer layer.
Quick definition
An attention head is one parallel attention mechanism inside a transformer layer.
- Category: Models
- Focus: model capability and fit
- Used in: Choosing a model that fits latency and cost constraints.
What it means
Heads specialize in patterns such as locality or long-range links. In models workflows, attention head often shapes model capability and fit.
How it works
Model architecture and scale determine capability. Context length, parameter count, and modality support vary across models.
Why it matters
Model architecture affects capability, context length, and speed.
Common use cases
- Choosing a model that fits latency and cost constraints.
- Selecting longer context for document-heavy workflows.
- Using specialized models for code, vision, or speech.
Example
A head may connect opening and closing brackets.
Pitfalls and tips
Bigger is not always better. Match the model to the task and evaluate in production.
In BoltAI
In BoltAI, this shows up in model selection and configuration.
