Fix TPU Spawn gather (#6896)

CHANGELOG.md

@@ -231,6 +231,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Fixed `EarlyStopping` logic when `min_epochs` or `min_steps` requirement is not met ([#6705](https://github.com/PyTorchLightning/pytorch-lightning/pull/6705))
 
 
+- Fixed TPU Spawn all gather ([#6896](https://github.com/PyTorchLightning/pytorch-lightning/pull/6896))
+
+
 - Fixed `--gpus` default for parser returned by `Trainer.add_argparse_args` ([#6898](https://github.com/PyTorchLightning/pytorch-lightning/pull/6898))
 
 

pytorch_lightning/accelerators/tpu.py

@@ -11,9 +11,8 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-from typing import Any, Callable, Optional, TYPE_CHECKING, Union
+from typing import Any, Callable, TYPE_CHECKING, Union
 
-import torch
 from torch.optim import Optimizer
 
 from pytorch_lightning.accelerators.accelerator import Accelerator
@@ -57,21 +56,6 @@ class TPUAccelerator(Accelerator):
     ) -> None:
         xm.optimizer_step(optimizer, barrier=False, optimizer_args={'closure': lambda_closure, **kwargs})
 
-    def all_gather(self, tensor: torch.Tensor, group: Optional[Any] = None, sync_grads: bool = False) -> torch.Tensor:
-        """
-        Function to gather a tensor from several distributed processes
-        Args:
-            tensor: tensor of shape (batch, ...)
-            group: not available with TPUs
-            sync_grads: not available with TPUs
-        Return:
-            A tensor of shape (world_size, batch, ...)
-        """
-        # todo: Add support for backward with all_gather
-        if isinstance(self.training_type_plugin, TPUSpawnPlugin) and self.training_type_plugin.is_distributed:
-            return xm.all_gather(tensor).view(-1, *tensor.shape)
-        return tensor
-
     def clip_gradients(self, optimizer: Optimizer, clip_val: Union[float, int], norm_type: float = 2.0):
 
         model = self.lightning_module

pytorch_lightning/plugins/training_type/tpu_spawn.py

@@ -195,14 +195,14 @@ class TPUSpawnPlugin(DDPSpawnPlugin):
         return obj
 
     def reduce_boolean_decision(self, decision: bool) -> bool:
-        decision = torch.tensor(int(decision), device=self.device)
-        decision = self.reduce(decision, "sum")
+        decision = torch.tensor(int(decision), device=self.lightning_module.device)
+        decision = self.reduce(decision, reduce_op="sum")
         decision = bool(decision == self.world_size)
         return decision
 
     def reduce(self, output, group: Optional[Any] = None, reduce_op: Optional[Union[ReduceOp, str]] = None):
         if not isinstance(output, torch.Tensor):
-            output = torch.tensor(output, device=self.device)
+            output = torch.tensor(output, device=self.lightning_module.device)
 
         _invalid_reduce_op = isinstance(reduce_op, ReduceOp) and reduce_op != ReduceOp.SUM
         _invalid_reduce_op_str = isinstance(reduce_op, str) and reduce_op.lower() not in ("sum", "mean", "avg")
@@ -267,3 +267,15 @@ class TPUSpawnPlugin(DDPSpawnPlugin):
         if _OMEGACONF_AVAILABLE:
             checkpoint = apply_to_collection(checkpoint, (DictConfig, ListConfig), OmegaConf.to_container)
         self.save({k: v for k, v in checkpoint.items() if k != "callbacks"}, filepath)
+
+    def all_gather(self, tensor: torch.Tensor, group: Optional[Any] = None, sync_grads: bool = False) -> torch.Tensor:
+        """
+        Function to gather a tensor from several distributed processes
+        Args:
+            tensor: tensor of shape (batch, ...)
+            group: not available with TPUs
+            sync_grads: not available with TPUs
+        Return:
+            A tensor of shape (world_size, batch, ...)
+        """
+        return xm.all_gather(tensor.unsqueeze(0))

tests/models/test_tpu.py

@@ -229,8 +229,8 @@ def test_tpu_clip_grad_by_value(tmpdir):
         progress_bar_refresh_rate=0,
         max_epochs=4,
         tpu_cores=1,
-        limit_train_batches=4,
-        limit_val_batches=4,
+        limit_train_batches=10,
+        limit_val_batches=10,
         gradient_clip_val=0.5,
         gradient_clip_algorithm='value'
     )