Open Models Almanac
Methodsmodel architecturetoken mixersoftmax attention

single mention · model architecture

Strict vision-language padding mask

Masks left-padded vision-language tokens so they do not contribute to the action head.

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

Also called strict_vl_padding_mask.

Filed under model architecture › token mixer › softmax attention