00:20:49	Rohit Gandikota:	Your speaker is not working?
00:33:09	Vishwanathan Ayyappan Kasi:	May I know if the padding tokens were used to unify the token length or for any other ablation studies?
00:40:45	Vishwanathan Ayyappan Kasi:	As we see a darker band along the column for “1 representative token” does that imply, only one of the token more important ie holds the complete notion/summary of sentence ?
00:45:22	Vishwanathan Ayyappan Kasi:	Will there be a Q/A session ?
00:54:39	Vishwanathan Ayyappan Kasi:	For the token rank analysis, why didn’t we analyse the attentions layers but the residual and MLP ?