Can hopper support recent published 1D scaling of FP8 in cuBlasLt

Accelerated Computing GPU-Accelerated Libraries
1

Not sure if this is supported for Hopper already or in future road map?

https://docs.nvidia.com/deeplearning/transformer-engine/user-guide/api/common.html#transformer_engine.common.recipe.DelayedScaling

Thanks
Gino

2

As I understand, TE 2.0 only supports mx scaling for Blackwell, but not for Hopper. And currently the best practice of FP8 for Hopper is Deepseek V3’s group-wise scaling, then would future TE release support group-wise scaling for Hopper, instead of per-tensor scaling?