Hi everyone
I profiled a simple python file that simulates the self attention mechanism in transformers . Sometimes the achieved kernel point in roofline diagram is between two horizontal lines, actually above the line of computation boundary. What does it mean? Has anyone experienced it before?
Could you share a screenshot and/or your NCU report to better understand what is going on?
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.