Attention Maps Visualization

How can I get the attention weights from an open_clip model? The clip library had the option to allow attention weights by setting need_weigths to True, and then the attention weights are returned in model.encode_image. How can this be done with this library?