Cannot specify GPU to train the model, only GPU 0 can be used

[lfbcccc]

Describe the bug
Cannot specify GPU to train the model, only GPU 0 can be used

Reproduction

1. What command or script did you run?

train.py

3. Did you make any modifications on the code or config? Did you understand what you have modified?

Only when I change the Gpu_id to 1 will an error occur

RuntimeError: Expected tensor for 'out' to have the same device as tensor for argument #3 'batch2'; but device 1 does not equal 0 (while checking arguments for baddbmm)

[xiexinch]

Please refer to https://mmsegmentation.readthedocs.io/en/latest/user_guides/4_train_test.html#launch-multiple-jobs-on-a-single-machine

[lfbcccc]

When I use the following command

CUDA_VISIBLE_DEVICES=1 sh tools/dist_train.sh local_configs/segnext/tiny/segnext.tiny.512x512.ade.160k.py 1 --work-dir output/voc_coal_dilation_gam_big_1024_80k_3_gai

This error may occur.
train.py: error: unrecognized arguments: local_configs/segnext/tiny/segnext.tiny.512x512.ade.160k.py
Killing subprocess 172132
How can I resolve it
please

[xiexinch]

Did you make any modifications to the train.py script?

[lfbcccc]

Yes, modifications were made during single card training, mainly including

In addition, I would like to inquire if there is a method for directly specifying GPU training in train.py

[wujiang0156]

Please refer to https://mmsegmentation.readthedocs.io/en/latest/user_guides/4_train_test.html#launch-multiple-jobs-on-a-single-machine

How to achieve single-machine multi-Gpus training on windows？
train models on win10 with multi-Gpus??