How can we get attention weights from example sequence and structure?

[what-is-what]

How can we get attention weights from example sequence and structure? There were no arguments to get attention weights in transformer blocks, unlike esm2.