LtNvfp4Matmul‘s scale is NULL!

[AnnaTrainingG]

When testing the LtNvfp4Matmul example on an SM_101 architecture GPU, I encountered the following runtime error:
"[cublasLt][702814][Error][cublasLtMatmulAlgoGetHeuristic] The scaling mode for A scale (VEC16_UE4M3) requires a custom pointer for A scale
cuBLAS API failed with status 7"

I noticed that the a_scale parameter being passed is NULL. Could you please advise how to modify this?

[rsdubtso]

Cannot reproduce locally with cuBLAS 12.9.

Can you please provide more info about cuBLAS and driver version? Did you make any modifications to the source?