dummy master_host in distributed.init()

zhijian-liu / torchpack

A neural network training interface based on PyTorch, with a focus on flexibility

https://pypi.org/project/torchpack/

MIT License

61 stars 15 forks source link

dummy master_host in distributed.init() #20

Closed RalphMao closed 3 years ago

RalphMao commented 3 years ago

PR 19 has the issue when actually calling the DistributedDataParallel module. Calling torch.distributed.init_process_group with a dummy master_host can solve this issue.