A few questions about the training process

[offchan42]

I am wondering about how you guys code, debug, and train this model.

1. Are you writing code and debugging the model using local machine or cloud machine? For example, if our GPU is not big enough maybe we'll need to use something like Lightning Studio for cloud development.
2. I want to know how to replicate the entire training from scratch and how costly it is. So what is the training set used to train this model? Can it be disclosed?
3. I am assuming that the architecture for StableCascade and Stable Diffusion 1.5 is different enough that it requires re-training all the extensions such as ControlNet or Lora. Do you need to re-train the ControlNet for this model?
4. Is the training code provided in this repository roughly the same one that you used for training StableCascade here?

Thanks. Your effort coding this model is greatly appreciated!