Auto3DSeg reduce GPU memory & num_workers

[Djennifer]

Hi, I am trying to train Auto3DSeg using the VerSe dataset. The GPU's available to me have 8GB and I believe that a larger one is needed to run Auto3DSeg. Now my training just stops after the first few epochs. However, I was wondering if there are ways to reduce the memory needed on the GPU. I already downsampled and cropped my data and this is enough for nnU-Net.

Another question is regarding the customizability of certain parameters. I noticed when training a U-Net with MONAI and Pytorch Lightning I needed to reduce the num_workers to 0 or 1. In Auto3DSeg I see that there are 8 workers on the train dataloader. Can I reduce this parameter? And if so how?