Zero ablate hook

Zero ablate hook.

zero_ablate_hook(value, hook)

Zero ablate hook.

Parameters:

Name	Type	Description	Default
value	Tensor	The activations to store.	required
hook	HookPoint	The hook point.	required

Example

dummy_hook_point = HookPoint() value = torch.ones(2, 3) zero_ablate_hook(value, dummy_hook_point) tensor([[0., 0., 0.], [0., 0., 0.]])

Returns:

Type	Description
Tensor	Replaced activations.

Source code in sparse_autoencoder/source_model/zero_ablate_hook.py

def zero_ablate_hook(
    value: Tensor,
    hook: HookPoint,  # noqa: ARG001
) -> Tensor:
    """Zero ablate hook.

    Args:
        value: The activations to store.
        hook: The hook point.

    Example:
        >>> dummy_hook_point = HookPoint()
        >>> value = torch.ones(2, 3)
        >>> zero_ablate_hook(value, dummy_hook_point)
        tensor([[0., 0., 0.],
                [0., 0., 0.]])

    Returns:
        Replaced activations.
    """
    return torch.zeros_like(value)