neuron_explainer/activations/derived_scalars/attention.py [739:757]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    )  # derive scalar of shape (n_tokens, n_attended_to_tokens, d_model)
    autoencoder_scalar_deriver = make_autoencoder_latent_scalar_deriver_factory(
        autoencoder_node_type
    )(
        dst_config
    )  # derive scalar of shape (n_tokens, n_latents)

    assert dst_config.layer_indices is not None
    assert len(dst_config.layer_indices) == 1
    layer_index = dst_config.layer_indices[0]

    autoencoder_context = dst_config.get_autoencoder_context(autoencoder_node_type)
    assert autoencoder_context is not None
    pre_act_encoder_derivative = make_autoencoder_pre_act_encoder_derivative(
        autoencoder_context, layer_index
    )
    activation_fn_derivative = make_autoencoder_activation_fn_derivative(
        autoencoder_context, layer_index
    )
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



neuron_explainer/activations/derived_scalars/attention.py [835:854]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    )  # derive scalar of shape (n_tokens, n_heads, v_channels)

    autoencoder_scalar_deriver = make_autoencoder_latent_scalar_deriver_factory(
        autoencoder_node_type
    )(
        dst_config
    )  # derive scalar of shape (n_tokens, n_latents)

    assert dst_config.layer_indices is not None
    assert len(dst_config.layer_indices) == 1
    layer_index = dst_config.layer_indices[0]

    autoencoder_context = dst_config.get_autoencoder_context(autoencoder_node_type)
    assert autoencoder_context is not None
    pre_act_encoder_derivative = make_autoencoder_pre_act_encoder_derivative(
        autoencoder_context, layer_index
    )
    activation_fn_derivative = make_autoencoder_activation_fn_derivative(
        autoencoder_context, layer_index
    )
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -