Error with DistributedSampler

[vickyli99]

Hi, I am trying to run your code but I have the issue with "Default process group has not been initialized, "
RuntimeError: Default process group has not been initialized, please make sure to call init_process_group.
I checked the initialization of DistribuedSampler and it looks fine to me, not sure what went wrong. Any help will be highly appreciated!

[bosung]

Hi, thanks for bringing this issue up. Have you tried -m torch.distributed.launch? You can use like this python -m torch.distributed.launch run_bert_multitask.py [options].

[yys-1]

Hi, I am trying to run your code but I have the issue with "Default process group has not been initialized, " RuntimeError: Default process group has not been initialized, please make sure to call init_process_group. I checked the initialization of DistribuedSampler and it looks fine to me, not sure what went wrong. Any help will be highly appreciated请问您解决了吗，多卡训练的如何用单卡跑