towhee.models.mdmmt.bert_mmt

Logic for the Transformer architecture used for MMT.

Classes

BertAttention	Self-attention layer.
BertEmbeddings	Construct the embeddings from word, position and token_type embeddings.
BertEncoder	Complete Bert Model (Transformer encoder).
BertIntermediate	Fully-connected layer, part 1.
BertLayer	Complete Bert layer.
BertMMT	Bert Model.
BertOutput	Fully-connected layer, part 2.
BertPooler	Extraction of a single output embedding.
BertSelfAttention	Self-attention mechanism.
BertSelfOutput	Self-attention output.