Missing “local-rank” when Training Classification from Scratch

[Jaakk0F]

Dear InternImage Developer:

Thanks for reading my message. I encountered the following error when I trained the classification model in ImageNet1K.

InternImage training and evaluation script: error: the following arguments are required: --local-rank

The script is modified from the [Training from Scratch on ImageNet-1K]. In my local machine, there are 2 GPUs. (https://github.com/OpenGVLab/InternImage/blob/master/classification/README.md#training-from-scratch-on-imagenet-1k).

python -m torch.distributed.launch --nproc_per_node 2 --master_port 12345  main.py --cfg configs/without_lr_decay/internimage_t_1k_224.yaml --data-path /path/to/data

I follow the install guidance in the classification README.

When I disabled the required condition for the "local_rank" argument. The script is stuck in constructing ModelEma when running with multiple GPUs. But the script can run with a single GPU with nproc_per_node as 1.

[EasonXiao-888]

I also encountered this problem. Have you solved it?

[SpiderJack0516]

@Jaakk0F I have same question.