Model techniques map
Techniquesmodel architecturetoken mixersoftmax attentiongrouped-query attention

specific method · filed under model architecture

Multi-query attention

An attention method described here as performing core attention in an MQA manner, including an MQA mode of MLA.

Also called Multi-Query Attention (MQA), Multi-Query Attention (MQA) mode of MLA.

sources
2
models
3
lab adopt it
1
strongest
used

How sources treat it

One count per evidence span, weakest treatment to strongest.

mentioned 1used 1

Documented in

Evidence

2 spans quoted from the sources, strongest treatment first.

CSA then performs core attention in a Multi-Query Attention (MQA) manner

usedmodel architecturein DeepSeek-V4DeepSeek

In the future, we plan to migrate the inference backend to the training engine and uniformly adopt the Multi-Query Attention (MQA) mode of MLA for inference

mentionedmodel architecturein GLM-5Z.ai

Filed alongside

Other methods under model architecture :: token mixer :: softmax attention :: grouped-query attention.