Transformers and LLMs

Cross-Attention, explained

Cross-attention lets one sequence attend to information from a different sequence or modality.
Back to Glossary

Part of the Semantic Notion technical glossary.

Short Definition

Cross-attention lets one sequence attend to information from a different sequence or modality.

Intuition

One representation asks questions of another representation, such as generated text looking back at an image or source sentence.

Technical Definition

Cross-attention usually forms queries from one stream and keys and values from another stream, enabling conditional information flow across inputs.

Example

An image captioning model may use text queries that attend to visual features.

Common Misunderstandings

Cross-attention is not the same as concatenating inputs.
It is a mechanism used inside many architectures, not only translation models.

Same Category

More from Transformers and LLMs

AttentionAttention is a neural network mechanism that lets a model decide which parts of an input are most relevant when producing a representation or output.Context WindowA context window is the maximum amount of input and generated tokens a model can consider at one time.Hybrid AttentionHybrid attention combines different attention patterns, such as local and global attention, inside one model.KV CacheA KV cache stores attention keys and values from previous tokens so autoregressive generation does not recompute them every step.Multimodal ModelA multimodal model can process or generate more than one kind of data, such as text, images, audio, or video.Positional EncodingPositional encoding gives a transformer information about token order and position.

Related Reading

Essays that use this concept

Transformers and LLMsSelf-Attention vs Cross-Attention in Transformers

Start here

Need the broader concept map?

Return to the glossary index to move from this definition into adjacent AI, ML, mathematics, and computer science terms.

Browse Glossary