Fix edgecase for early stopping

src/lightning/pytorch/loops/fit_loop.py

@@ -182,6 +182,11 @@ def done(self) -> bool:
             rank_zero_info(f"`Trainer.fit` stopped: `max_steps={self.max_steps!r}` reached.")
             return True
 
+        # Check early stopping before max_epochs to prioritize user-initiated stopping
+        if self.trainer.should_stop and self._can_stop_early:
+            rank_zero_debug("`Trainer.fit` stopped: `trainer.should_stop` was set.")
+            return True
+
         # `processed` is increased before `on_train_epoch_end`, the hook where checkpoints are typically saved.
         # we use it here because the checkpoint data won't have `completed` increased yet
         assert isinstance(self.max_epochs, int)
@@ -192,10 +197,6 @@ def done(self) -> bool:
             rank_zero_info(f"`Trainer.fit` stopped: `max_epochs={self.max_epochs!r}` reached.")
             return True
 
-        if self.trainer.should_stop and self._can_stop_early:
-            rank_zero_debug("`Trainer.fit` stopped: `trainer.should_stop` was set.")
-            return True
-
         return False
 
     @property

tests/tests_pytorch/callbacks/test_early_stopping.py

@@ -509,6 +509,51 @@ def test_early_stopping_log_info(log_rank_zero_only, world_size, global_rank, ex
         log_mock.assert_not_called()
 
 
+def test_early_stopping_message_priority_over_max_epochs(caplog):
+    """Test that early stopping message takes priority over max_epochs message when both conditions are met."""
+
+    class TestModel(BoringModel):
+        def __init__(self):
+            super().__init__()
+            self.validation_called = False
+
+        def validation_step(self, batch, batch_idx):
+            result = super().validation_step(batch, batch_idx)
+            self.log("monitor_metric", 1000.0 if not self.validation_called else 999.0)
+            self.validation_called = True
+            return result
+
+        def on_validation_end(self):
+            if self.current_epoch == 0:
+                self.trainer.should_stop = True
+
+    model = TestModel()
+    early_stopping = EarlyStopping(
+        monitor="monitor_metric",
+        mode="min",
+        patience=0,
+        verbose=True,
+    )
+
+    trainer = Trainer(
+        max_epochs=1,
+        callbacks=[early_stopping],
+        enable_progress_bar=False,
+        enable_model_summary=False,
+        logger=False,
+        limit_train_batches=1,
+        limit_val_batches=1,
+    )
+
+    with caplog.at_level(logging.DEBUG, logger="lightning.pytorch.utilities.rank_zero"):
+        trainer.fit(model)
+
+    assert "`Trainer.fit` stopped: `trainer.should_stop` was set." in caplog.text
+    assert trainer.should_stop is True
+    assert early_stopping.stopped_epoch >= 0
+    assert "`Trainer.fit` stopped: `max_epochs=1` reached." not in caplog.text
+
+
 class ModelWithHighLoss(BoringModel):
     def on_validation_epoch_end(self):
         self.log("val_loss", 10.0)

tests/tests_pytorch/loops/test_training_loop.py

@@ -166,24 +166,27 @@ def test_fit_loop_done_log_messages(caplog):
     fit_loop.epoch_progress.current.processed = 3
     fit_loop.max_epochs = 3
     trainer.should_stop = True
-    assert fit_loop.done
-    assert "max_epochs=3` reached" in caplog.text
-    caplog.clear()
-    fit_loop.max_epochs = 5
-
     fit_loop.epoch_loop.min_steps = 0
     with caplog.at_level(level=logging.DEBUG, logger="lightning.pytorch.utilities.rank_zero"):
         assert fit_loop.done
     assert "should_stop` was set" in caplog.text
+    caplog.clear()
 
+    trainer.should_stop = False
+    assert fit_loop.done
+    assert "max_epochs=3` reached" in caplog.text
+    caplog.clear()
+    fit_loop.max_epochs = 5
+
+    trainer.should_stop = True
     fit_loop.epoch_loop.min_steps = 100
     assert not fit_loop.done
 
 
 @pytest.mark.parametrize(
     ("min_epochs", "min_steps", "current_epoch", "early_stop", "fit_loop_done", "raise_debug_msg"),
     [
-        (4, None, 100, True, True, False),
+        (4, None, 100, True, True, True),
         (4, None, 3, False, False, False),
         (4, 10, 3, False, False, False),
         (None, 10, 4, True, True, True),