How do I run the training script if I only have one graphics card?

[billfjj]

How do I run the training script if I only have one graphics card?

[billfjj]

I tried, but reported an error

[lkeab]

hi, you can modify the nproc_per_node number from 8 to 1 here

[ymq2017]

This looks like a torch version problem. Different versions have different definitions between local_rank and local-rank. You could try one of these methods.

1. Use a lower version of PyTorch
2. Use torchrun to replace python -m torch.distributed.launch

[billfjj]

This is my version of PyTorch, pytorch==1.13.0 torchvision==0.14.0 torchaudio==0.13.0 pytorch-cuda=11.7. But it still gives an error

Is it because of Windows? Have you ever trained on Windows?

[vishakhalall]

I fixed this issue by finding the correct compatible version of pytorch, found the CUDA version using nvcc --version and found the compatible version of PyTorch https://pytorch.org/get-started/previous-versions/