Loss go NAN when running classification model of Seaformer_B

[shiyutang]

I run the following command as described in the readme, but the loss goes NAN. I wonder why this happens.

[wwqq]

Try to train the model with 8 gpus.

[shiyutang]

Thanks a lot, I must forget to change to 8 after test.

[Lun Zhongwang]

hi，I trained the model with 8 gpus, but also goes nan.

The --resume arg is not specified in the startup command, will this not affect it?

[Lun Zhongwang]

Thanks a lot, I must forget to change to 8 after test.

Have you solved this problem?

[shiyutang]

Yes, It needs to be trained with 8 gpus.

[Lun Zhongwang]

Yes, It needs to be trained with 8 gpus.

Thanks.