use parallel loader

Author: lezwon

pytorch_lightning/trainer/evaluation_loop.py

@@ -135,6 +135,7 @@
 from pytorch_lightning.utilities import rank_zero_warn
 
 try:
+    import torch_xla
     import torch_xla.distributed.parallel_loader as xla_pl
     import torch_xla.core.xla_model as xm
 except ImportError:
@@ -249,8 +250,8 @@ def _evaluate(self, model: LightningModule, dataloaders, max_batches: int, test_
             dl_outputs = []
 
             # on TPU we have to wrap it under the ParallelLoader
-            if self.use_tpu and self.tpu_id is None:
-                device = xm.xla_device()
+            if self.use_tpu:
+                device = torch_xla._XLAC._xla_get_default_device()
                 dataloader = xla_pl.ParallelLoader(dataloader, [device])
                 dataloader = dataloader.per_device_loader(device)
 

pytorch_lightning/trainer/training_loop.py

@@ -167,6 +167,7 @@ def training_step(self, batch, batch_idx):
     APEX_AVAILABLE = True
 
 try:
+    import torch_xla
     import torch_xla.distributed.parallel_loader as xla_pl
     import torch_xla.core.xla_model as xm
 except ImportError:
@@ -412,8 +413,8 @@ def run_training_epoch(self):
         train_dataloader = self.train_dataloader
 
         # on TPU we have to wrap it under the ParallelLoader
-        if self.use_tpu and self.tpu_id is None:
-            device = xm.xla_device()
+        if self.use_tpu:
+            device = torch_xla._XLAC._xla_get_default_device()
             train_dataloader = xla_pl.ParallelLoader(train_dataloader, [device])
             train_dataloader = train_dataloader.per_device_loader(device)