remove dataloader patching on the LightningModule (#9764)

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: rohitgr7 <[email protected]>
Co-authored-by: Carlos Mocholí <[email protected]>

Author: 4 people

CHANGELOG.md

@@ -513,6 +513,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Removed `process_idx` from the `{DDPSpawnPlugin,TPUSpawnPlugin}.new_process` methods ([#10022](https://github.com/PyTorchLightning/pytorch-lightning/pull/10022))
 
 
+- Removed automatic patching of `{train,val,test,predict}_dataloader()` on the `LightningModule` ([#9764](https://github.com/PyTorchLightning/pytorch-lightning/pull/9764))
+
+
 ### Fixed
 
 
@@ -594,6 +597,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 
 
+- Fixed undesired side effects being caused by `Trainer` patching dataloader methods on the `LightningModule` ([#9764](https://github.com/PyTorchLightning/pytorch-lightning/pull/9764))
+
+
 ## [1.4.9] - 2021-09-30
 
 - Fixed `lr_find` to generate same results on multiple calls ([#9704](https://github.com/PyTorchLightning/pytorch-lightning/pull/9704))

pl_examples/basic_examples/autoencoder.py

@@ -181,8 +181,8 @@ def cli_main():
         trainer_defaults={"callbacks": ImageSampler(), "max_epochs": 10},
     )
     cli.trainer.fit(cli.model, datamodule=cli.datamodule)
-    cli.trainer.test(ckpt_path="best")
-    predictions = cli.trainer.predict(ckpt_path="best")
+    cli.trainer.test(ckpt_path="best", datamodule=cli.datamodule)
+    predictions = cli.trainer.predict(ckpt_path="best", datamodule=cli.datamodule)
     print(predictions[0])
 
 

pl_examples/basic_examples/backbone_image_classifier.py

@@ -124,8 +124,8 @@ def predict_dataloader(self):
 def cli_main():
     cli = LightningCLI(LitClassifier, MyDataModule, seed_everything_default=1234, save_config_overwrite=True, run=False)
     cli.trainer.fit(cli.model, datamodule=cli.datamodule)
-    cli.trainer.test(ckpt_path="best")
-    predictions = cli.trainer.predict(ckpt_path="best")
+    cli.trainer.test(ckpt_path="best", datamodule=cli.datamodule)
+    predictions = cli.trainer.predict(ckpt_path="best", datamodule=cli.datamodule)
     print(predictions[0])
 
 

pl_examples/basic_examples/dali_image_classifier.py

@@ -194,7 +194,7 @@ def cli_main():
 
     cli = LightningCLI(LitClassifier, MyDataModule, seed_everything_default=1234, save_config_overwrite=True, run=False)
     cli.trainer.fit(cli.model, datamodule=cli.datamodule)
-    cli.trainer.test(ckpt_path="best")
+    cli.trainer.test(ckpt_path="best", datamodule=cli.datamodule)
 
 
 if __name__ == "__main__":

pl_examples/basic_examples/simple_image_classifier.py

@@ -74,7 +74,7 @@ def cli_main():
         LitClassifier, MNISTDataModule, seed_everything_default=1234, save_config_overwrite=True, run=False
     )
     cli.trainer.fit(cli.model, datamodule=cli.datamodule)
-    cli.trainer.test(ckpt_path="best")
+    cli.trainer.test(ckpt_path="best", datamodule=cli.datamodule)
 
 
 if __name__ == "__main__":

pytorch_lightning/plugins/plugins_registry.py

@@ -127,11 +127,7 @@ def is_register_plugins_overridden(plugin: type) -> bool:
     else:
         return False
 
-    if hasattr(plugin_attr, "patch_loader_code"):
-        is_overridden = plugin_attr.patch_loader_code != str(super_attr.__code__)
-    else:
-        is_overridden = plugin_attr.__code__ is not super_attr.__code__
-    return is_overridden
+    return plugin_attr.__code__ is not super_attr.__code__
 
 
 def call_training_type_register_plugins(root: Path, base_module: str) -> None:

pytorch_lightning/plugins/training_type/deepspeed.py

@@ -628,8 +628,9 @@ def _auto_select_batch_size(self):
         # train_micro_batch_size_per_gpu is used for throughput logging purposes
         # by default we try to use the batch size of the loader
         batch_size = 1
-        if hasattr(self.lightning_module, "train_dataloader"):
-            train_dataloader = self.lightning_module.train_dataloader()
+        train_dl_source = self.lightning_module.trainer.data_connector._train_dataloader_source
+        if train_dl_source.is_defined():
+            train_dataloader = train_dl_source.dataloader()
             if hasattr(train_dataloader, "batch_sampler"):
                 batch_size = train_dataloader.batch_sampler.batch_size
         return batch_size

pytorch_lightning/plugins/training_type/tpu_spawn.py

@@ -20,14 +20,13 @@
 
 import torch
 import torch.multiprocessing as mp
-from torch.nn import Module
 from torch.utils.data import DataLoader
 
 import pytorch_lightning as pl
 from pytorch_lightning.overrides import LightningDistributedModule
 from pytorch_lightning.plugins.io.checkpoint_plugin import CheckpointIO
 from pytorch_lightning.plugins.training_type.ddp_spawn import DDPSpawnPlugin
-from pytorch_lightning.trainer.connectors.data_connector import _PatchDataLoader
+from pytorch_lightning.trainer.connectors.data_connector import DataConnector
 from pytorch_lightning.trainer.states import TrainerFn
 from pytorch_lightning.utilities import (
     _OMEGACONF_AVAILABLE,
@@ -96,19 +95,18 @@ def _validate_dataloader(dataloaders: Union[List[DataLoader], DataLoader]) -> No
                 )
 
     @staticmethod
-    def _validate_patched_dataloaders(model: Module) -> None:
+    def _validate_patched_dataloaders(model: "pl.LightningModule") -> None:
         """Validate and fail fast if the dataloaders were passed directly to fit."""
-        if hasattr(model, "train_dataloader") and isinstance(model.train_dataloader, _PatchDataLoader):
-            TPUSpawnPlugin._validate_dataloader(model.train_dataloader.dataloader)
-
-        if hasattr(model, "val_dataloader") and isinstance(model.val_dataloader, _PatchDataLoader):
-            TPUSpawnPlugin._validate_dataloader(model.val_dataloader.dataloader)
-
-        if hasattr(model, "test_dataloader") and isinstance(model.test_dataloader, _PatchDataLoader):
-            TPUSpawnPlugin._validate_dataloader(model.test_dataloader.dataloader)
-
-        if hasattr(model, "predict_dataloader") and isinstance(model.predict_dataloader, _PatchDataLoader):
-            TPUSpawnPlugin._validate_dataloader(model.predict_dataloader.dataloader)
+        connector: DataConnector = model.trainer.data_connector
+        sources = (
+            connector._train_dataloader_source,
+            connector._val_dataloader_source,
+            connector._test_dataloader_source,
+            connector._predict_dataloader_source,
+        )
+        for source in sources:
+            if not source.is_module():
+                TPUSpawnPlugin._validate_dataloader(source.instance)
 
     def connect(self, model: "pl.LightningModule") -> None:
         TPUSpawnPlugin._validate_patched_dataloaders(model)

pytorch_lightning/trainer/configuration_validator.py

@@ -38,7 +38,7 @@ def verify_loop_configurations(trainer: "pl.Trainer", model: "pl.LightningModule
     elif trainer.state.fn == TrainerFn.TESTING:
         __verify_eval_loop_configuration(model, "test")
     elif trainer.state.fn == TrainerFn.PREDICTING:
-        __verify_predict_loop_configuration(model)
+        __verify_predict_loop_configuration(trainer, model)
     __verify_dp_batch_transfer_support(trainer, model)
     _check_add_get_queue(model)
     # TODO(@daniellepintz): Delete _check_progress_bar in v1.7
@@ -65,7 +65,7 @@ def __verify_train_loop_configuration(trainer: "pl.Trainer", model: "pl.Lightnin
     # -----------------------------------
     # verify model has a train dataloader
     # -----------------------------------
-    has_train_dataloader = is_overridden("train_dataloader", model)
+    has_train_dataloader = trainer.data_connector._train_dataloader_source.is_defined()
     if not has_train_dataloader:
         raise MisconfigurationException(
             "No `train_dataloader()` method defined. Lightning `Trainer` expects as minimum a"
@@ -175,8 +175,8 @@ def __verify_eval_loop_configuration(model: "pl.LightningModule", stage: str) ->
         )
 
 
-def __verify_predict_loop_configuration(model: "pl.LightningModule") -> None:
-    has_predict_dataloader = is_overridden("predict_dataloader", model)
+def __verify_predict_loop_configuration(trainer: "pl.Trainer", model: "pl.LightningModule") -> None:
+    has_predict_dataloader = trainer.data_connector._predict_dataloader_source.is_defined()
     if not has_predict_dataloader:
         raise MisconfigurationException("Dataloader not found for `Trainer.predict`")
     # ----------------------------------------------

pytorch_lightning/trainer/connectors/data_connector.py

@@ -12,8 +12,9 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import os
+from dataclasses import dataclass
 from functools import partial
-from typing import Callable, Iterable, Optional, Union
+from typing import Iterable, Optional, Union
 
 import pytorch_lightning as pl
 from pytorch_lightning.utilities import rank_zero_deprecation
@@ -47,6 +48,11 @@ def __init__(
         self.test_data_fetcher = test_data_fetcher
         self.sanity_check_data_fetcher: Optional[AbstractDataFetcher] = None
 
+        self._train_dataloader_source = _DataLoaderSource(None, "")
+        self._val_dataloader_source = _DataLoaderSource(None, "")
+        self._test_dataloader_source = _DataLoaderSource(None, "")
+        self._predict_dataloader_source = _DataLoaderSource(None, "")
+
     @property
     def evaluation_data_fetcher(self) -> Optional[AbstractDataFetcher]:
         if self.trainer.sanity_checking:
@@ -190,27 +196,23 @@ def attach_dataloaders(
         test_dataloaders: Optional[EVAL_DATALOADERS] = None,
         predict_dataloaders: Optional[EVAL_DATALOADERS] = None,
     ) -> None:
-        # when dataloader is passed via fit, patch the train_dataloader
-        # functions to overwrite with these implementations
-        if train_dataloaders is not None:
-            self.trainer.train_dataloader = None
-            train_dataloader = _PatchDataLoader(train_dataloaders, "train")
-            train_dataloader.patch(model)
-
-        if val_dataloaders is not None:
-            self.trainer.val_dataloaders = None
-            val_dataloader = _PatchDataLoader(val_dataloaders, "val")
-            val_dataloader.patch(model)
-
-        if test_dataloaders is not None:
-            self.trainer.test_dataloaders = None
-            test_dataloader = _PatchDataLoader(test_dataloaders, "test")
-            test_dataloader.patch(model)
-
-        if predict_dataloaders is not None:
-            self.trainer.predict_dataloaders = None
-            predict_dataloader = _PatchDataLoader(predict_dataloaders, "predict")
-            predict_dataloader.patch(model)
+        self.trainer.train_dataloader = None
+        self.trainer.val_dataloaders = None
+        self.trainer.test_dataloaders = None
+        self.trainer.predict_dataloaders = None
+
+        self._train_dataloader_source = _DataLoaderSource(
+            train_dataloaders if train_dataloaders is not None else model, "train_dataloader"
+        )
+        self._val_dataloader_source = _DataLoaderSource(
+            val_dataloaders if val_dataloaders is not None else model, "val_dataloader"
+        )
+        self._test_dataloader_source = _DataLoaderSource(
+            test_dataloaders if test_dataloaders is not None else model, "test_dataloader"
+        )
+        self._predict_dataloader_source = _DataLoaderSource(
+            predict_dataloaders if predict_dataloaders is not None else model, "predict_dataloader"
+        )
 
     def attach_datamodule(
         self, model: "pl.LightningModule", datamodule: Optional["pl.LightningDataModule"] = None
@@ -219,11 +221,10 @@ def attach_datamodule(
         if datamodule is None:
             return
 
-        # Override loader hooks
-        dl_methods = ("train_dataloader", "val_dataloader", "test_dataloader", "predict_dataloader")
-        for method in dl_methods:
-            if is_overridden(method, datamodule):
-                setattr(model, method, getattr(datamodule, method))
+        self._train_dataloader_source = _DataLoaderSource(datamodule, "train_dataloader")
+        self._val_dataloader_source = _DataLoaderSource(datamodule, "val_dataloader")
+        self._test_dataloader_source = _DataLoaderSource(datamodule, "test_dataloader")
+        self._predict_dataloader_source = _DataLoaderSource(datamodule, "predict_dataloader")
 
         # Override data transfer hooks if dataset-specific to_device logic has been defined in datamodule
         batch_transfer_hooks = ("on_before_batch_transfer", "transfer_batch_to_device", "on_after_batch_transfer")
@@ -238,13 +239,6 @@ def attach_datamodule(
         if hasattr(datamodule, "data_pipeline"):
             model.data_pipeline = datamodule.data_pipeline
 
-    @staticmethod
-    def detach_data(model: "pl.LightningModule") -> None:
-        for stage in ("train", "val", "test", "predict"):
-            loader = getattr(model, f"{stage}_dataloader", None)
-            if isinstance(loader, _PatchDataLoader):
-                loader.unpatch(model)
-
     def teardown(self) -> None:
         if self.train_data_fetcher:
             self.train_data_fetcher.teardown()
@@ -260,32 +254,56 @@ def teardown(self) -> None:
             self.sanity_check_data_fetcher = None
 
 
-class _PatchDataLoader:
-    r"""
-    Callable object for patching dataloaders passed into trainer.fit().
-    Use this class to override model.*_dataloader() and be pickle-compatible.
+@dataclass
+class _DataLoaderSource:
+    """Stores the information where the dataloaders come from.
+
+    The source can be
 
-    Args:
-        dataloader: Dataloader object to return when called.
+    1. from a ``*_datalaoder()`` method on the :class:`~pytorch_lightning.core.lightning.LightningModule`,
+    2. from a ``*_datalaoder()`` method on the :class:`~pytorch_lightning.core.datamodule.LightningDataModule`,
+    3. a direct instance of a :class:`~torch.utils.data.DataLoader` or supported collections thereof.
+
+    Arguments:
+        instance: A LightningModule, LightningDataModule, or (a collection of) dataloader(s).
+        name: A name for this dataloader source. If the instance is a module, the name corresponds to the hook
+            that returns the desired dataloader(s).
     """
 
-    def __init__(self, dataloader: Union[TRAIN_DATALOADERS, EVAL_DATALOADERS], stage: str) -> None:
-        self.dataloader = dataloader
+    instance: Optional[Union[TRAIN_DATALOADERS, EVAL_DATALOADERS, "pl.LightningModule", "pl.LightningDataModule"]]
+    name: str
+
+    def dataloader(self) -> Union[TRAIN_DATALOADERS, EVAL_DATALOADERS]:
+        """Returns the dataloader from the source.
+
+        If the source is a module, the method with the corresponding :attr:`name` gets called.
+        """
+        from pytorch_lightning import LightningDataModule, LightningModule  # prevent cyclic import
+
+        if not self.name:
+            return self.instance
+
+        if isinstance(self.instance, LightningModule):
+            return self.instance.trainer.call_hook(self.name, pl_module=self.instance)
+
+        if isinstance(self.instance, LightningDataModule):
+            method = getattr(self.instance, self.name)
+            return method()
+
+        return self.instance
+
+    def is_defined(self) -> bool:
+        """Returns whether the source dataloader can be retrieved or not.
 
-        # cannot pickle __code__ so cannot verify if PatchDataloader
-        # exists which shows dataloader methods have been overwritten.
-        # so, we hack it by using the string representation
-        self.patch_loader_code = str(self.__call__.__code__)
-        self._old_loader: Optional[Callable] = None
-        self.stage = stage
+        If the source is a module it checks that the method with given :attr:`name` is overridden.
+        """
+        return not self.is_module() or is_overridden(self.name, self.instance)
 
-    def __call__(self) -> Union[TRAIN_DATALOADERS, EVAL_DATALOADERS]:
-        return self.dataloader
+    def is_module(self) -> bool:
+        """Returns whether the the DataLoader source is a LightningModule or a LightningDataModule.
 
-    def patch(self, model: "pl.LightningModule") -> None:
-        self._old_loader = getattr(model, self.stage + "_dataloader")
-        setattr(model, self.stage + "_dataloader", self)
+        It does not check whether ``*_dataloader`` methods are actually overridden.
+        """
+        from pytorch_lightning import LightningDataModule, LightningModule  # prevent cyclic import
 
-    def unpatch(self, model: "pl.LightningModule") -> None:
-        setattr(model, self.stage + "_dataloader", self._old_loader)
-        self._old_loader = None
+        return isinstance(self.instance, (LightningModule, LightningDataModule))