no_sync if paddle support else nullcontext

deepspeech/exps/deepspeech2/model.py

@@ -87,7 +87,8 @@ class DeepSpeech2Trainer(Trainer):
             # Disable gradient synchronizations across DDP processes.
             # Within this context, gradients will be accumulated on module
             # variables, which will later be synchronized.
-            context = self.model.no_sync
+            context = self.model.no_sync if (hasattr(self.model, "no_sync") and
+                                             self.parallel) else nullcontext
         else:
             # Used for single gpu training and DDP gradient synchronization
             # processes.

deepspeech/exps/u2/model.py

@@ -106,7 +106,8 @@ class U2Trainer(Trainer):
             # Within this context, gradients will be accumulated on module
             # variables, which will later be synchronized.
             # When using cpu w/o DDP, model does not have `no_sync`
-            context = self.model.no_sync if self.parallel else nullcontext
+            context = self.model.no_sync if (hasattr(self.model, "no_sync") and
+                                             self.parallel) else nullcontext
         else:
             # Used for single gpu training and DDP gradient synchronization
             # processes.

deepspeech/exps/u2_kaldi/model.py

@@ -105,7 +105,8 @@ class U2Trainer(Trainer):
             # Disable gradient synchronizations across DDP processes.
             # Within this context, gradients will be accumulated on module
             # variables, which will later be synchronized.
-            context = self.model.no_sync
+            context = self.model.no_sync if (hasattr(self.model, "no_sync") and
+                                             self.parallel) else nullcontext
         else:
             # Used for single gpu training and DDP gradient synchronization
             # processes.

deepspeech/exps/u2_st/model.py

@@ -110,7 +110,8 @@ class U2STTrainer(Trainer):
             # Disable gradient synchronizations across DDP processes.
             # Within this context, gradients will be accumulated on module
             # variables, which will later be synchronized.
-            context = self.model.no_sync
+            context = self.model.no_sync if (hasattr(self.model, "no_sync") and
+                                             self.parallel) else nullcontext
         else:
             # Used for single gpu training and DDP gradient synchronization
             # processes.