Longformer(LED) map out Global Tokens

[anwarika]

Is there a way to see what the model is using for global tokens in summarization?