Method forward

modelscope/models/multi_modal/mgeo/backbone.py:572–638

(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_value=None,
        output_attentions=False,
    )

        self.output = BertOutput(config)

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_value=None,
        output_attentions=False,
    ):
        # decoder uni-directional self-attention cached key/values tuple is at
        # positions 1,2
        self_attn_past_key_value = past_key_value[:
                                                  2] if past_key_value is not None else None
        self_attention_outputs = self.attention(
            hidden_states,
            attention_mask,
            head_mask,
            output_attentions=output_attentions,
            past_key_value=self_attn_past_key_value,
        )
        attention_output = self_attention_outputs[0]

        outputs = self_attention_outputs[1:-1]
        present_key_value = self_attention_outputs[-1]

        if self.has_cross_attention:
            assert encoder_hidden_states is not None, 'encoder_hidden_states must be given for cross-attention layers'

            if type(encoder_hidden_states) == list:
                cross_attention_outputs = self.crossattention(
                    attention_output,
                    attention_mask,
                    head_mask,
                    encoder_hidden_states[(self.layer_num
                                           - self.config.fusion_layer)
                                          % len(encoder_hidden_states)],
                    encoder_attention_mask[(self.layer_num
                                            - self.config.fusion_layer)
                                           % len(encoder_hidden_states)],
                    output_attentions=output_attentions,
                )
                attention_output = cross_attention_outputs[0]
                outputs = outputs + cross_attention_outputs[1:-1]

            else:
                cross_attention_outputs = self.crossattention(
                    attention_output,
                    attention_mask,
                    head_mask,
                    encoder_hidden_states,
                    encoder_attention_mask,
                    output_attentions=output_attentions,
                )
                attention_output = cross_attention_outputs[0]
                outputs = outputs + cross_attention_outputs[
                    1:
                    -1]  # add cross attentions if we output attention weights