The setting of batchsize and training time

[pushi]

Thank you for your excellent work! We are tried to reproduce your results, but we note that the batchsize for a single GPU is very small (i.e., 4). Could you explain why you use so small batchsize on an A100 GPU? Was the Stable diffusion model trained for about 24 days on 256 A100 GPUs？

Could you provide the configuration file and related codes (including the dataloader and the 10% dropping of the text-condition) for training the model? we note that there is only an inference configuration file here.

[mia01]

@pushi did you find the answers to this question? would like to know as well