How to run multiple instances at once?

[roseodimm]

I have 4 GPUs in the same PC, and after I put set CUDA_VISIBLE_DEVICES=0-3 in the new run1-3.bat that I create, the software opens with the correct GPU, but it only allows me to train on one GPU and forces me to close another three instances.

How to make it work like A1111 or Forge so that you can run and generate separate tasks per instance?

I already tried the original kohya_ss, but I can't get the multi-GPU accelerator training to work, so I think it is better to train 4 LoRA on 4 GPUs.

[Jelosus2]

I'm not aware if that's possible or not because I don't have 2+ GPUs but yeah probably your best bet is to train 1 LoRA per GPU

[derrian-distro]

technically, it's not designed with multi-gpu support in mind, because 99% of people won't have more than one GPU, but you can likely get it to run multiple instances by changing the launch port in the backend config file between each launch, it would be kind of tedious, but that might work, but since I don't have more than one GPU, I cannot confirm this will work