RuntimeError: Default process group has not been initialized, please make sure to call init_process_

Problems encountered when using mmsegmentation framework:

 File "C:\software\Anaconda3\envs\python36\lib\site-packages\torch\distributed\distributed_c10d.py", line 347, in _get_default_group
    raise RuntimeError("Default process group has not been initialized, "
RuntimeError: Default process group has not been initialized, please make sure to call init_process_group.

After debugging and positioning, it is found that there is a problem with the normalization of a convolution module:

        self.linear_fuse = ConvModule(
            in_channels=embedding_dim*4,
            out_channels=embedding_dim,
            kernel_size=1,
            norm_cfg=dict(type='SyncBN', requires_grad=True)
          
        )

Norm here_ In CFG, if it is multi card training, use “syncbn”; If it is a single card training, change the type to ‘BN’.

Read More: