Add LightningCLI(run=False|True)

CHANGELOG.md

@@ -27,6 +27,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Added `log_graph` argument for `watch` method of `WandbLogger` ([#8662](https://github.com/PyTorchLightning/pytorch-lightning/pull/8662))
 
 
+- Added `LightningCLI(run=False|True)` to choose whether to run a `Trainer` subcommand ([#8751](https://github.com/PyTorchLightning/pytorch-lightning/pull/8751))
+
+
 - Fault-tolerant training:
     * Added `FastForwardSampler` and `CaptureIterableDataset` injection to data loading utilities ([#8366](https://github.com/PyTorchLightning/pytorch-lightning/pull/8366))
 

docs/source/common/lightning_cli.rst

@@ -260,17 +260,30 @@ file. Loading a defaults file :code:`my_cli_defaults.yaml` in the current workin
 
 .. testcode::
 
-    cli = LightningCLI(
-        MyModel,
-        MyDataModule,
-        parser_kwargs={"default_config_files": ["my_cli_defaults.yaml"]},
-    )
+    cli = LightningCLI(MyModel, MyDataModule, parser_kwargs={"default_config_files": ["my_cli_defaults.yaml"]})
 
 To load a file in the user's home directory would be just changing to :code:`~/.my_cli_defaults.yaml`. Note that this
 setting is given through :code:`parser_kwargs`. More parameters are supported. For details see the `ArgumentParser API
 <https://jsonargparse.readthedocs.io/en/stable/#jsonargparse.core.ArgumentParser.__init__>`_ documentation.
 
 
+Instantiation only mode
+^^^^^^^^^^^^^^^^^^^^^^^
+
+The CLI is designed to start fitting without minimal code changes. On class instantiation, the CLI will automatically
+call ``trainer.fit(...)`` internally so you don't have to do it. To avoid this, you can set the following argument:
+
+.. testcode::
+
+    cli = LightningCLI(MyModel, run=False)  # True by default
+    # you'll have to call fit yourself:
+    cli.trainer.fit(cli.model)
+
+
+This can be useful to implement custom logic without having to subclass the CLI, but still using the CLI's instantiation
+and argument parsing capabilities.
+
+
 Trainer Callbacks and arguments with class type
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 

pytorch_lightning/utilities/cli.py

@@ -213,6 +213,7 @@ def __init__(
         parser_kwargs: Dict[str, Any] = None,
         subclass_mode_model: bool = False,
         subclass_mode_data: bool = False,
+        run: bool = True,
     ) -> None:
         """
         Receives as input pytorch-lightning classes (or callables which return pytorch-lightning classes), which are
@@ -259,6 +260,8 @@ def __init__(
             subclass_mode_data: Whether datamodule can be any `subclass
                 <https://jsonargparse.readthedocs.io/en/stable/#class-type-and-sub-classes>`_
                 of the given class.
+            run: Whether subcommands should be added to run a :class:`~pytorch_lightning.trainer.trainer.Trainer`
+                method. If set to ``False``, the trainer and model classes will be instantiated only.
         """
         self.model_class = model_class
         self.datamodule_class = datamodule_class
@@ -283,10 +286,12 @@ def __init__(
         self.before_instantiate_classes()
         self.instantiate_classes()
         self.add_configure_optimizers_method_to_model()
-        self.prepare_fit_kwargs()
-        self.before_fit()
-        self.fit()
-        self.after_fit()
+
+        if run:
+            self.prepare_fit_kwargs()
+            self.before_fit()
+            self.fit()
+            self.after_fit()
 
     def init_parser(self) -> None:
         """Method that instantiates the argument parser"""

tests/utilities/test_cli.py

@@ -687,3 +687,12 @@ def __init__(self, optim1: dict, optim2: dict, scheduler: dict):
     assert isinstance(cli.model.optim1, torch.optim.Adam)
     assert isinstance(cli.model.optim2, torch.optim.SGD)
     assert isinstance(cli.model.scheduler, torch.optim.lr_scheduler.ExponentialLR)
+
+
+@pytest.mark.parametrize("run", (False, True))
+def test_lightning_cli_disabled_run(run):
+    with mock.patch("sys.argv", ["any.py"]), mock.patch("pytorch_lightning.Trainer.fit") as fit_mock:
+        cli = LightningCLI(BoringModel, run=run)
+    fit_mock.assert_called() if run else fit_mock.assert_not_called()
+    assert isinstance(cli.trainer, Trainer)
+    assert isinstance(cli.model, LightningModule)