Fix LightningCLI failing when both module and data module save hyperparameters

src/lightning/pytorch/CHANGELOG.md

@@ -5,6 +5,21 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 
+## [unreleased] - YYYY-MM-DD
+
+### Added
+
+### Changed
+
+- Merging of hparams when logging now ignores parameter names that begin with underscore `_` ([#20221](https://github.com/Lightning-AI/pytorch-lightning/pull/20221))
+
+### Removed
+
+### Fixed
+
+- Fix LightningCLI failing when both module and data module save hyperparameters due to conflicting internal `_class_path` parameter ([#20221](https://github.com/Lightning-AI/pytorch-lightning/pull/20221))
+
+
 ## [2.4.0] - 2024-08-06
 
 ### Added

src/lightning/pytorch/loggers/utilities.py

@@ -69,6 +69,9 @@ def _log_hyperparams(trainer: "pl.Trainer") -> None:
         lightning_hparams = pl_module.hparams_initial
         inconsistent_keys = []
         for key in lightning_hparams.keys() & datamodule_hparams.keys():
+            if key.startswith("_"):
+                # Prevent merging LightningCLI's internal hparams and discourage of private param names
+                continue
             lm_val, dm_val = lightning_hparams[key], datamodule_hparams[key]
             if (
                 type(lm_val) != type(dm_val)

tests/tests_pytorch/test_cli.py

@@ -952,6 +952,29 @@ def test_lightning_cli_save_hyperparameters_untyped_module(cleandir):
     assert model.kwargs == {"x": 1}
 
 
+class TestDataSaveHparams(BoringDataModule):
+    def __init__(self, batch_size: int = 32, num_workers: int = 4):
+        super().__init__()
+        self.save_hyperparameters()
+        self.batch_size = batch_size
+        self.num_workers = num_workers
+
+
+def test_lightning_cli_save_hyperparameters_merge(cleandir):
+    config = {
+        "model": {
+            "class_path": f"{__name__}.TestModelSaveHparams",
+        },
+        "data": {
+            "class_path": f"{__name__}.TestDataSaveHparams",
+        },
+    }
+    with mock.patch("sys.argv", ["any.py", "fit", f"--config={json.dumps(config)}", "--trainer.max_epochs=1"]):
+        cli = LightningCLI(auto_configure_optimizers=False)
+    assert set(cli.model.hparams) == {"optimizer", "scheduler", "activation", "_instantiator", "_class_path"}
+    assert set(cli.datamodule.hparams) == {"batch_size", "num_workers", "_instantiator", "_class_path"}
+
+
 @pytest.mark.parametrize("fn", [fn.value for fn in TrainerFn])
 def test_lightning_cli_trainer_fn(fn):
     class TestCLI(LightningCLI):