Extracting attention weights of summarization model

Hello, I am working on a summarization model with BERT and am keen on getting additional data (attention weight) generated from the output text (generated summary). thanks