espnet2.lid.conditioned_layer.Lang2VecConditionedLayer

source

class espnet2.lid.conditioned_layer.Lang2VecConditionedLayer(layer, encoder, pooling, projector, prediction, conditioning_layer=None, detach_prediction=False, detach_condition=False)

Bases: Module

Predict lang2vec after an upstream layer and add its projected residual.

The original layer owns attention and checkpointing. The LID model consumes intermediate_out for the auxiliary loss and clears it after each call. As with self-conditioned interCTC, the model creates the projection, which may be shared across several wrappers.

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(hidden_states, attention_mask=None, output_attentions=False, **kwargs)

Preserve the upstream output tuple and pass attention arguments through.