[Question] Why hybrid embedding is bound with async data reading?

[jayz0123]

Hi HugeCTR team,

Currently I am doing a performance breakdown of DLRM, where I want to test out the performance of asynchronous data reader and hybrid embedding. However, I found that in HugeCTR, RawAsync type must be used together with the hybrid embedding feature.

I am wondering is there any reason for this? And is there any way to split them apart?