About batch_size of the second training stage---training CFW

[YangGangZhiQi]

Hi, thanks for this great works. I am trying to training stableSR in my local dataset. When training CFW, I have a question about the batch size in autoencoder_kl_64x64x4_resi.yaml. I noticed that the batch size of the first training stage was 6 and the second stage was 1. Why the batch size of the second training stage is only one? Could I set it more than one?

I would appreciate it if anyone could help me.