RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method

[sapkun]

Hello, when i train my own data with the sript train_instruct_pix2pix_sdxl.py with --dataloader_num_workers=2, it gives me this error:
RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method.
But If I change the --dataloader_num_workers=0, it will work. How to run this script with multiple dataloader_num_workers?
Can anyone help in this regard? thank you!

[sayakpaul]

You can refer to https://github.com/sayakpaul/instructpix2pix-sdxl for a more dedicated treatment of that script to run on multiple GPUs.