[Fix] Ensure we set the eval/train flag correctly on accelerator model (#6877)

* Ensure we move the model to eval mode before running evaluation

* Ensure we set the flag appropriately across all stages

* Add test, move hooks logic

* Apply same fix to the validate loop

* Update pytorch_lightning/trainer/trainer.py

* Fix function name

* Fix order, add predict

* Shorten the name

* Fix input dm, drop duplicate on predict start hook call, as it's called in the setup function

* Use hook, remove double call

pytorch_lightning/core/hooks.py

@@ -114,13 +114,13 @@ class ModelHooks:
         """
         Sets the model to eval during the val loop
         """
-        self.eval()
+        self.trainer.model.eval()
 
     def on_validation_model_train(self) -> None:
         """
         Sets the model to train during the val loop
         """
-        self.train()
+        self.trainer.model.train()
 
     def on_validation_batch_start(self, batch: Any, batch_idx: int, dataloader_idx: int) -> None:
         """
@@ -172,19 +172,19 @@ class ModelHooks:
         """
         Sets the model to train during the test loop
         """
-        self.train()
+        self.trainer.model.train()
 
     def on_test_model_eval(self) -> None:
         """
         Sets the model to eval during the test loop
         """
-        self.eval()
+        self.trainer.model.eval()
 
     def on_predict_model_eval(self) -> None:
         """
         Sets the model to eval during the predict loop
         """
-        self.eval()
+        self.trainer.model.eval()
 
     def on_epoch_start(self) -> None:
         """

pytorch_lightning/trainer/predict_loop.py

@@ -44,7 +44,6 @@ class PredictLoop(object):
         model_ref.on_predict_model_eval()
 
     def setup(self, model, max_batches, dataloaders):
-        self.trainer.call_hook("on_predict_start")
 
         # copy properties for forward overrides
         self.trainer.model_connector.copy_trainer_model_properties(model)

pytorch_lightning/trainer/trainer.py

@@ -582,11 +582,11 @@ class Trainer(
         self.checkpoint_connector.has_trained = False
 
         # enable train mode
-        model = self.lightning_module
-        model.train()
+        self.model.train()
         torch.set_grad_enabled(True)
 
         # reload data when needed
+        model = self.lightning_module
         self.train_loop.reset_train_val_dataloaders(model)
 
         # hook
@@ -772,8 +772,6 @@ class Trainer(
         return eval_loop_results
 
     def run_predict(self):
-        self.predict_loop.on_predict_start()
-
         # prepare dataloaders
         dataloaders, max_batches = self.predict_loop.get_predict_dataloaders()
 
@@ -789,6 +787,9 @@ class Trainer(
         model.zero_grad()
         torch.set_grad_enabled(False)
 
+        # call hook
+        self.predict_loop.on_predict_start()
+
         # set up the eval loop
         self.predict_loop.setup(model, max_batches, dataloaders)
 

tests/trainer/test_trainer.py

@@ -1438,7 +1438,9 @@ def test_trainer_setup_call(tmpdir, stage):
 )
 @patch("pytorch_lightning.loggers.tensorboard.TensorBoardLogger.log_metrics")
 def test_log_every_n_steps(log_metrics_mock, tmpdir, train_batches, max_steps, log_interval):
+
     class TestModel(BoringModel):
+
         def training_step(self, *args, **kwargs):
             self.log("foo", -1)
             return super().training_step(*args, **kwargs)
@@ -1888,3 +1890,33 @@ def test_exception_when_testing_or_validating_with_fast_dev_run(tmpdir):
         trainer.validate()
     with pytest.raises(MisconfigurationException, match=r"\.test\(\)` with `fast_dev_run=True"):
         trainer.test()
+
+
+class TrainerStagesModel(BoringModel):
+
+    def on_train_start(self) -> None:
+        assert self.trainer.model.training
+        assert self.training
+
+    def on_validation_start(self) -> None:
+        assert not self.trainer.model.training
+        assert not self.training
+
+    def on_test_start(self) -> None:
+        assert not self.trainer.model.training
+        assert not self.training
+
+    def on_predict_start(self) -> None:
+        assert not self.trainer.model.training
+        assert not self.training
+
+
+@pytest.mark.parametrize(['accelerator', 'num_processes'],
+                         [(None, 1), pytest.param('ddp', 2, marks=RunIf(skip_windows=True))])
+def test_model_in_correct_mode_during_stages(tmpdir, accelerator, num_processes):
+    model = TrainerStagesModel()
+    trainer = Trainer(default_root_dir=tmpdir, accelerator=accelerator, num_processes=num_processes, fast_dev_run=True)
+    trainer.fit(model)
+    trainer.validate(model)
+    trainer.test(model)
+    trainer.predict(model, model.val_dataloader())