single mention · model architecture
Shape attention mask for action-prefix conditioning
Replaces a causal mask with a shape-specific mask for action-prefix conditioning; the evidence does not identify the shape name beyond its displayed fragment.
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 a -shape attention mask.
Filed under model architecture › token mixer › softmax attention