This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHubThe 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
| Parameter | Data Type | Required | Range | Description |
|---|---|---|---|---|
model | MODEL | Yes | - | The UNet model to modify with attention scaling factors |
q | FLOAT | No | 0.0 - 10.0 | Multiplication factor for query component (default: 1.0) |
k | FLOAT | No | 0.0 - 10.0 | Multiplication factor for key component (default: 1.0) |
v | FLOAT | No | 0.0 - 10.0 | Multiplication factor for value component (default: 1.0) |
out | FLOAT | No | 0.0 - 10.0 | Multiplication factor for output component (default: 1.0) |
Outputs
| Output Name | Data Type | Description |
|---|---|---|
MODEL | MODEL | The modified UNet model with scaled attention components |