Swin Transformer training and evaluation script: error: the following arguments are required: --local_rank

[hailuoS]

Hello, I used the following command to train
"python -m torch.distributed.launch --nproc_per_node 8 --master_port 12345 main.py --cfg configs/swin/swin_tiny_patch4_window7_224.yaml --batch-size 128 --fused_window_process --fused_layernorm --optim=fused_adam --data-path /home/datasets/imagenet"
but got "Swin Transformer training and evaluation script: error: the following arguments are required: --local_rank"，how can I fix this problem, thanks

[LvQiangWen]

I have the same error, sad. Have you solved this problem?

[BitCalSaul]

hi @hailuoS @LvQiangWen did you find the answer

[wjyshuai]

I have resolved this issue and have conducted distributed training tests on both PyTorch 1.8 and 1.10 versions. I mainly made some code deletions to main.py and config.py. Specific:
'''python
Delete lines 72-73 of code from main.py
Delete lines 341-344 of code from config.py, then add config.LOCAL-RANK=int (os. environment ['LOCAL-RANK '])
'''
The above is my solution, I hope it will be useful to you.

[wjyshuai]

hi @hailuoS @LvQiangWen did you find the answer

You can try my method.

[BitCalSaul]

hi @hailuoS @LvQiangWen did you find the answer

You can try my method.

Bro, thanks for your reply. i solved this issue but still appreciate it :)