tensor2tensor/models/evolved_transformer.py [376:395]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
                hidden_state,
                encoder_output,
                encoder_decoder_attention_bias,
                hparams.attention_key_channels or hparams.hidden_size,
                hparams.attention_value_channels or hparams.hidden_size,
                hparams.hidden_size,
                hparams.num_heads,
                hparams.attention_dropout,
                max_relative_position=hparams.max_relative_position,
                heads_share_relative_embedding=(
                    hparams.heads_share_relative_embedding),
                add_relative_to_values=hparams.add_relative_to_values,
                save_weights_to=save_weights_to,
                cache=attention_cache,
                make_image_summary=make_image_summary,
                dropout_broadcast_dims=attention_dropout_broadcast_dims,
                max_length=hparams.get("max_length"),
                vars_3d=hparams.get("attention_variables_3d"),
                activation_dtype=hparams.get("activation_dtype", "float32"),
                weight_dtype=hparams.get("weight_dtype", "float32"))
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



tensor2tensor/models/evolved_transformer.py [585:604]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
                hidden_state,
                encoder_output,
                encoder_decoder_attention_bias,
                hparams.attention_key_channels or hparams.hidden_size,
                hparams.attention_value_channels or hparams.hidden_size,
                hparams.hidden_size,
                hparams.num_heads,
                hparams.attention_dropout,
                max_relative_position=hparams.max_relative_position,
                heads_share_relative_embedding=(
                    hparams.heads_share_relative_embedding),
                add_relative_to_values=hparams.add_relative_to_values,
                save_weights_to=save_weights_to,
                cache=attention_cache,
                make_image_summary=make_image_summary,
                dropout_broadcast_dims=attention_dropout_broadcast_dims,
                max_length=hparams.get("max_length"),
                vars_3d=hparams.get("attention_variables_3d"),
                activation_dtype=hparams.get("activation_dtype", "float32"),
                weight_dtype=hparams.get("weight_dtype", "float32"))
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -