Open Models Almanac
Methodsmodel architecturetoken mixersoftmax attentionglobal attention

single mention · model architecture

Full attention with multi-head attention for reliability

Full attention specifically described as using multi-head attention for reliability.

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

Also called Multi-Head Attention (MHA) for reliability.

Filed under model architecture › token mixer › softmax attention › global attention