Hi,
Thanks for your open-source project.
I noticed that your training was performed on 2 A100 (80G) for 200 epochs in README, which are very expensive.
I would like to know approximately how much GPU memory is required for the training and whether 3090/4090 can meet your requirements.