Comment by maestroquirk

Need one for vision models too tbh. Token/s doesn't really map easily