transformer_lens.components.layer_norm_pre

Hooked Transformer Layer Norm Pre Component.

This module contains all the component LayerNormPre.

class transformer_lens.components.layer_norm_pre.LayerNormPre(cfg: Union[Dict, HookedTransformerConfig])

Bases: Module

__init__(cfg: Union[Dict, HookedTransformerConfig])

LayerNormPre - the ‘center and normalise’ part of LayerNorm. Length is normally d_model, but is d_mlp for softmax. Not needed as a parameter. This should only be used in inference mode after folding in LayerNorm weights

forward(x: Union[Float[Tensor, 'batch pos d_model'], Float[Tensor, 'batch pos head_index d_model']]) → Union[Float[Tensor, 'batch pos d_model'], Float[Tensor, 'batch pos head_index d_model']]

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.