Skip to main content
This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub
The UNetSelfAttentionMultiply node applies multiplication factors to the query, key, value, and output components of the self-attention mechanism in a UNet model. It allows you to scale different parts of the attention computation to experiment with how attention weights affect the model’s behavior.

Inputs

ParameterData TypeRequiredRangeDescription
modelMODELYes-The UNet model to modify with attention scaling factors
qFLOATNo0.0 - 10.0Multiplication factor for query component (default: 1.0)
kFLOATNo0.0 - 10.0Multiplication factor for key component (default: 1.0)
vFLOATNo0.0 - 10.0Multiplication factor for value component (default: 1.0)
outFLOATNo0.0 - 10.0Multiplication factor for output component (default: 1.0)

Outputs

Output NameData TypeDescription
MODELMODELThe modified UNet model with scaled attention components