Apply sc.pp.highly_variable_genes to SCT-normalized residuals or counts

Can the scanpy team answer this questions?
To integrate multiple datasets, I have applied SCT-normalization followed by harmony using Seurat. Then, I applied scanpy for the downstream analysis. After filtering out unwanted population, I want to selected HVG and redo the clustering. Now, the questions is that sc.experimental.pp.highly_variable_genes() expect raw counts (integer) which is not my case; while sc.pp.highly_variable_genes() has three flavor [[Satija15]], [[Zheng17]], [[Stuart19]]. Which one should I apply?

Thank you,