finished dist (#911)

pytorch_lightning/trainer/data_loading.py

@@ -25,6 +25,15 @@ try:
 except ImportError:
     APEX_AVAILABLE = False
 
+try:
+    import torch_xla
+    import torch_xla.core.xla_model as xm
+    import torch_xla.distributed.xla_multiprocessing as xmp
+
+    XLA_AVAILABLE = True
+except ImportError:
+    XLA_AVAILABLE = False
+
 
 class TrainerDataLoadingMixin(ABC):
 
@@ -217,12 +226,15 @@ class TrainerDataLoadingMixin(ABC):
 
         # on TPUs load each dataloader only on process 0
         # this will trigger the data downloads
-        if self.use_tpu:
+        if self.use_tpu and XLA_AVAILABLE:
             if self.tpu_local_core_rank == 0:
                 self.get_train_dataloader()
                 self.get_test_dataloaders()
                 self.get_val_dataloaders()
 
+            # wait for all processes to catch up
+            torch_xla.core.xla_model.rendezvous()
+
         # support IterableDataset for train data
         self.is_iterable_train_dataloader = (
             EXIST_ITER_DATASET and isinstance(self.get_train_dataloader().dataset, IterableDataset))

pytorch_lightning/trainer/trainer.py

@@ -983,6 +983,11 @@ class Trainer(TrainerIOMixin,
         if self.use_ddp or self.use_ddp2:
             dist.barrier()
 
+        # wait for all models to restore weights
+        if self.on_tpu and XLA_AVAILABLE:
+            # wait for all processes to catch up
+            torch_xla.core.xla_model.rendezvous()
+
         # set up checkpoint callback
         self.configure_checkpoint_callback()
 

pytorch_lightning/trainer/training_io.py

@@ -106,6 +106,15 @@ from pytorch_lightning.overrides.data_parallel import (
     LightningDataParallel,
 )
 
+try:
+    import torch_xla
+    import torch_xla.core.xla_model as xm
+    import torch_xla.distributed.xla_multiprocessing as xmp
+
+    XLA_AVAILABLE = True
+except ImportError:
+    XLA_AVAILABLE = False
+
 
 class TrainerIOMixin(ABC):
 
@@ -125,6 +134,7 @@ class TrainerIOMixin(ABC):
         self.early_stop_callback = None
         self.lr_schedulers = None
         self.optimizers = None
+        self.on_tpu = None
         self.num_training_batches = None
         self.accumulate_grad_batches = None
 
@@ -170,6 +180,11 @@ class TrainerIOMixin(ABC):
             # wait for all processes to catch up
             dist.barrier()
 
+        # wait for all models to restore weights
+        if self.on_tpu and XLA_AVAILABLE:
+            # wait for all processes to catch up
+            torch_xla.core.xla_model.rendezvous()
+
         # clear cache after restore
         if self.on_gpu:
             torch.cuda.empty_cache()