Open Models Almanac
Methodsmodel architecturetoken mixersoftmax attentionmulti-head latent attention

implementation detail · filed under model architecture

MQA mode of MLA

The MLA mode in which each latent key-value vector is shared across query heads.

The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See multi-head latent attention for the methods filed beside it.

sources
2
model
1
lab adopt it
1
strongest
core

Documented in

Evidence

2 quotes from 2 sources, one card per model, each opening with its strongest.

How sources treat it
used 1core 1

DeepSeek-V3.2

DeepSeekadopted2 sources

Therefore, we implement DSA based on the MQA (Shazeer, 2019) mode of MLA, where each latent vector (the key-value entry of MLA) will be shared across all query heads of the query token.

coremodel architecturein DeepSeek-V3.2DeepSeek
1 more from this model

we implement DSA based on the MQA mode of MLA, where each latent vector (the key-value entry of MLA) will be shared across all query heads of the query token.

usedmodel architecturein DeepSeek-V3.2DeepSeek
↑