Hardware Requirements about gigagan-pytorch

[kaykyr]

Thank you guys! Thank you for sharing a bit of knowledge with me. I'll study imagen-pytorch.

[lucidrains]

@kaykyr thanks for the small sponsorship!

[axel578]

Just a quick note, no model are officially relaesed from this repo(and so have been trained fully and released), so you have to train from scratch, the computing requirements scales depending on the number of parameters you are creating for your model, a rtx 4090 is more than enough if you make a small model with small images, but if you choose bigger paramters in the gan architecture dimension(which means your gan will hold a bigger number of parameters), the memory requirements will scale accordingly, the best would be to start with a small model, take the example code provided, and experiment with the parameters, most of all the GigaGan parameters of the class in the code example will scale memory requirements up.

This repo is pretty new, you may want to use imagen-pytorch.

[lucidrains]

yea agreed with Axel. a lot of research papers have already came out using imagen-pytorch, while gigagan is still fairly new and no reported success training it yet