Fix lightning cli crashing when trainer defaults contain callback (#21192)

* add parent
* add unittesting of function
* add testing in cli
* add changelog
* split tests
* linter

---------

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

Author: 3 people

src/lightning/pytorch/CHANGELOG.md

@@ -37,8 +37,12 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Fixed edgecase when `max_trials` is reached in `Tuner.scale_batch_size` ([#21187](https://github.com/Lightning-AI/pytorch-lightning/pull/21187))
 
 
+- Fixed case where `LightningCLI` could not be initialized with `trainer_default` containing callbacks ([#21192](https://github.com/Lightning-AI/pytorch-lightning/pull/21192))
+
+
 - Fixed missing reset when `ModelPruning` is applied with lottery ticket hypothesis ([#21191](https://github.com/Lightning-AI/pytorch-lightning/pull/21191))
 
+
 ---
 
 ## [2.5.5] - 2025-09-05

src/lightning/pytorch/trainer/connectors/callback_connector.py

@@ -240,7 +240,7 @@ def _reorder_callbacks(callbacks: list[Callback]) -> list[Callback]:
 
 
 def _validate_callbacks_list(callbacks: list[Callback]) -> None:
-    stateful_callbacks = [cb for cb in callbacks if is_overridden("state_dict", instance=cb)]
+    stateful_callbacks = [cb for cb in callbacks if is_overridden("state_dict", instance=cb, parent=Callback)]
     seen_callbacks = set()
     for callback in stateful_callbacks:
         if callback.state_key in seen_callbacks:

tests/tests_pytorch/test_cli.py

@@ -1917,3 +1917,29 @@ def __init__(self, main_param: int = 1):
         cli = LightningCLI(MainModule, run=False, parser_kwargs={"parser_mode": "jsonnet"})
 
     assert cli.config["model"]["main_param"] == 2
+
+
+def test_lightning_cli_callback_trainer_default(cleandir):
+    """Check that callbacks passed as trainer_defaults are properly instantiated."""
+    with mock.patch("sys.argv", ["any.py"]):
+        cli = LightningCLI(
+            BoringModel,
+            BoringDataModule,
+            trainer_defaults={
+                "logger": {
+                    "class_path": "lightning.pytorch.loggers.TensorBoardLogger",
+                    "init_args": {
+                        "save_dir": ".",
+                        "name": "demo",
+                    },
+                },
+                "callbacks": {
+                    "class_path": "lightning.pytorch.callbacks.ModelCheckpoint",
+                    "init_args": {
+                        "monitor": "val_loss",
+                    },
+                },
+            },
+            run=False,
+        )
+    assert any(isinstance(c, ModelCheckpoint) for c in cli.trainer.callbacks)

tests/tests_pytorch/trainer/connectors/test_callback_connector.py

@@ -32,7 +32,7 @@
 )
 from lightning.pytorch.callbacks.batch_size_finder import BatchSizeFinder
 from lightning.pytorch.demos.boring_classes import BoringModel
-from lightning.pytorch.trainer.connectors.callback_connector import _CallbackConnector
+from lightning.pytorch.trainer.connectors.callback_connector import _CallbackConnector, _validate_callbacks_list
 
 
 @patch("lightning.pytorch.trainer.connectors.callback_connector._RICH_AVAILABLE", False)
@@ -104,12 +104,12 @@ def test_checkpoint_callbacks_are_last(tmp_path):
     ]
 
 
-class StatefulCallback0(Callback):
+class StatefulCallbackContent0(Callback):
     def state_dict(self):
         return {"content0": 0}
 
 
-class StatefulCallback1(Callback):
+class StatefulCallbackContent1(Callback):
     def __init__(self, unique=None, other=None):
         self._unique = unique
         self._other = other
@@ -126,9 +126,9 @@ def test_all_callback_states_saved_before_checkpoint_callback(tmp_path):
     """Test that all callback states get saved even if the ModelCheckpoint is not given as last and when there are
     multiple callbacks of the same type."""
 
-    callback0 = StatefulCallback0()
-    callback1 = StatefulCallback1(unique="one")
-    callback2 = StatefulCallback1(unique="two", other=2)
+    callback0 = StatefulCallbackContent0()
+    callback1 = StatefulCallbackContent1(unique="one")
+    callback2 = StatefulCallbackContent1(unique="two", other=2)
     checkpoint_callback = ModelCheckpoint(dirpath=tmp_path, filename="all_states")
     model = BoringModel()
     trainer = Trainer(
@@ -147,9 +147,9 @@ def test_all_callback_states_saved_before_checkpoint_callback(tmp_path):
     trainer.fit(model)
 
     ckpt = torch.load(str(tmp_path / "all_states.ckpt"), weights_only=True)
-    state0 = ckpt["callbacks"]["StatefulCallback0"]
-    state1 = ckpt["callbacks"]["StatefulCallback1{'unique': 'one'}"]
-    state2 = ckpt["callbacks"]["StatefulCallback1{'unique': 'two'}"]
+    state0 = ckpt["callbacks"]["StatefulCallbackContent0"]
+    state1 = ckpt["callbacks"]["StatefulCallbackContent1{'unique': 'one'}"]
+    state2 = ckpt["callbacks"]["StatefulCallbackContent1{'unique': 'two'}"]
     assert "content0" in state0
     assert state0["content0"] == 0
     assert "content1" in state1
@@ -323,3 +323,80 @@ def state_dict(self):
 
     with pytest.raises(RuntimeError, match="Found more than one stateful callback of type `MockCallback`"):
         Trainer(callbacks=[MockCallback(), MockCallback()])
+
+
+# Test with single stateful callback
+class StatefulCallback(Callback):
+    def state_dict(self):
+        return {"state": 1}
+
+
+# Test with multiple stateful callbacks with unique state keys
+class StatefulCallback1(Callback):
+    @property
+    def state_key(self):
+        return "unique_key_1"
+
+    def state_dict(self):
+        return {"state": 1}
+
+
+class StatefulCallback2(Callback):
+    @property
+    def state_key(self):
+        return "unique_key_2"
+
+    def state_dict(self):
+        return {"state": 2}
+
+
+@pytest.mark.parametrize(
+    ("callbacks"),
+    [
+        [Callback(), Callback()],
+        [StatefulCallback()],
+        [StatefulCallback1(), StatefulCallback2()],
+    ],
+)
+def test_validate_callbacks_list_function(callbacks: list):
+    """Test the _validate_callbacks_list function directly with various scenarios."""
+    _validate_callbacks_list(callbacks)
+
+
+# Test with multiple stateful callbacks with same state key
+class ConflictingCallback(Callback):
+    @property
+    def state_key(self):
+        return "same_key"
+
+    def state_dict(self):
+        return {"state": 1}
+
+
+# Test with different types of stateful callbacks that happen to have same state key
+class AnotherConflictingCallback(Callback):
+    @property
+    def state_key(self):
+        return "same_key"  # Same key as ConflictingCallback
+
+    def state_dict(self):
+        return {"state": 3}
+
+
+@pytest.mark.parametrize(
+    ("callbacks", "match_msg"),
+    [
+        (
+            [ConflictingCallback(), ConflictingCallback()],
+            "Found more than one stateful callback of type `ConflictingCallback`",
+        ),
+        (
+            [ConflictingCallback(), Callback(), ConflictingCallback()],
+            "Found more than one stateful callback of type `ConflictingCallback`",
+        ),
+        ([ConflictingCallback(), AnotherConflictingCallback()], "Found more than one stateful callback"),
+    ],
+)
+def test_raising_error_validate_callbacks_list_function(callbacks: list, match_msg: str):
+    with pytest.raises(RuntimeError, match=match_msg):
+        _validate_callbacks_list(callbacks)