feat: add tests for save_hyperparameters with ignore behavior and composition handling (#20718)

* add tests to check for `save_hyperparameter: ignore`
* update

---------

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

Author: deependujha

tests/tests_pytorch/models/test_hparams.py

@@ -440,6 +440,41 @@ def __init__(self, same_arg="parent_default", other_arg="other"):
     assert parent.child.hparams == {"same_arg": "cocofruit"}
 
 
+@pytest.mark.parametrize("base_class", [HyperparametersMixin, LightningModule, LightningDataModule])
+def test_save_hyperparameters_ignore(base_class):
+    """Test if `save_hyperparameter` applies the ignore list correctly during initialization."""
+
+    class PLSubclass(base_class):
+        def __init__(self, learning_rate=1e-3, optimizer="adam"):
+            super().__init__()
+            self.save_hyperparameters(ignore=["learning_rate"])
+
+    pl_instance = PLSubclass(learning_rate=0.01, optimizer="sgd")
+    assert pl_instance.hparams == {"optimizer": "sgd"}
+
+
+@pytest.mark.parametrize("base_class", [HyperparametersMixin, LightningModule, LightningDataModule])
+def test_save_hyperparameters_ignore_under_composition(base_class):
+    """Test that in a composed system, hyperparameter saving skips ignored fields from nested modules."""
+
+    class ChildModule(base_class):
+        def __init__(self, dropout, activation, init_method):
+            super().__init__()
+            self.save_hyperparameters(ignore=["dropout", "activation"])
+
+    class ParentModule(base_class):
+        def __init__(self, batch_size, optimizer):
+            super().__init__()
+            self.child = ChildModule(dropout=0.1, activation="relu", init_method="xavier")
+
+    class PipelineWrapper:  # not a Lightning subclass on purpose
+        def __init__(self, run_id="abc123", seed=42):
+            self.parent_module = ParentModule(batch_size=64, optimizer="adam")
+
+    pipeline = PipelineWrapper()
+    assert pipeline.parent_module.child.hparams == {"init_method": "xavier", "batch_size": 64, "optimizer": "adam"}
+
+
 class LocalVariableModelSuperLast(BoringModel):
     """This model has the super().__init__() call at the end."""