Recommend for large scale training dataset

[NNDam]

Hi, thanks for your amazing work. I've several question and hope you can give some advises.

• Can default arch & config work good with large scale training dataset, e.g 1M image
• Did you tried VQVAEGANMultiHeadTransformer and compared result with default VQVAEGAN, with my knowledge Transformer will work better with large scale dataset
• Some hyper parameters turning to deal with large scale dataset

Thanks again