BeyondAttention#

class txv.exp.BeyondAttention(model: Module)#

Link to Paper: Transformer Interpretability Beyond Attention Visualization

__init__(model: Module) → None#
Parameters:

model (torch.nn.Module) – A model from txv.vit

Caution

The model must be an LRP model. You can use the LRP version of a model by passing lrp=True in the model function.

explain(input: Tensor, index: int | None = None, alpha: float = 1.0, abm: bool = True, **kwargs) → Tensor#
Parameters:

  • input (torch.Tensor) – Input tensor

  • index (int, optional) – Index of the class to explain, by default the predicted class is explained

  • alpha (float, optional) – Alpha value for LRP, by default 1.0

  • abm (bool, optional) – Architecture based modification, by default True