[BLIP] fix cross attentions for BlipTextEncoder (#22515)

src/transformers/models/blip/modeling_blip_text.py

@@ -445,6 +445,7 @@ class BlipTextEncoder(nn.Module):
                 next_decoder_cache += (layer_outputs[-1],)
             if output_attentions:
                 all_self_attentions = all_self_attentions + (layer_outputs[1],)
+                all_cross_attentions = all_cross_attentions + (layer_outputs[2],)
 
         if output_hidden_states:
             all_hidden_states = all_hidden_states + (hidden_states,)