[QST]how to use 68_hopper_fp8_warp_specialized_grouped_gemm_with_blockwise_scaling

[youzjuer]

when the group number is 3: ([(1280, 1024),(512, 5120),(512, 256)], [(1280, 1024),(256, 5120),(512, 256)]), it can work well ,
however i change the input as :([(1280, 1024),(512, 5120),(512, 256),(512, 256)], [(1280, 1024),(256, 5120),(512, 256),(512, 256)]),
the result is :

，why？

[github-actions]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

[hwu36]

when the group number is 3: ([(1280, 1024),(512, 5120),(512, 256)], [(1280, 1024),(256, 5120),(512, 256)])

how can I interpret the problem sizes? In the first group size, m=1280, n=1280, k = 1024?