Mtp doesnt improve decode on rtx 5090
#5
by jasionkajakub - opened
i checked n1 to n3 mtp it is decrease decode form 68tps to 50 but it increase to 100tps on qwen 3.6 27b nvfp4 form nvidia
This comment has been hidden (marked as Off-Topic)
i checked n1 to n3 mtp it is decrease decode form 68tps to 50 but it increase to 100tps on qwen 3.6 27b nvfp4 form nvidia