Training details of SD-XL

[kkk3146]

Thank you for your exceptional work!

Your paper mentioned the distillation training process on SD-v1.5: "Training is conducted on 8 RTX6000Ada GPUs with a total batch size of 96 under 512x512 resolution for 100k iterations, requiring ~1 day to complete".

Could you provide additional information regarding the training process for SD-XL? More specifically, the number of iterations or epochs, the training time, and how to choose the final model.

We eagerly await your response.