ddp pickle

Author: williamFalcon

pytorch_lightning/trainer/data_loading.py

@@ -133,8 +133,8 @@ def auto_add_sampler(self, dataloader: DataLoader, train: bool) -> DataLoader:
                 world_size = {
                     'ddp': self.num_nodes * self.num_processes,
                     'ddp2': self.num_nodes,
+                    'ddp_cpu': self.num_processes * self.num_nodes
                 }
-                import pdb; pdb.set_trace()
                 sampler = DistributedSampler(
                     dataloader.dataset,
                     num_replicas=world_size.get(self.distributed_backend, 0),