feat: allow configurable join_char for WandbLogger metric formatting

src/lightning/pytorch/loggers/wandb.py

@@ -279,6 +279,7 @@ def any_lightning_module_function_or_hook(self):
         experiment: WandB experiment object. Automatically set when creating a run.
         checkpoint_name: Name of the model checkpoint artifact being logged.
         add_file_policy: If "mutable", copies file to tempdirectory before upload.
+        join_char: Separator character used to format metric keys before logging.
         \**kwargs: Arguments passed to :func:`wandb.init` like `entity`, `group`, `tags`, etc.
 
     Raises:
@@ -306,6 +307,7 @@ def __init__(
         prefix: str = "",
         checkpoint_name: Optional[str] = None,
         add_file_policy: Literal["mutable", "immutable"] = "mutable",
+        join_char: Optional[str] = None,
         **kwargs: Any,
     ) -> None:
         if not _WANDB_AVAILABLE:
@@ -351,6 +353,7 @@ def __init__(
         self._name = self._wandb_init.get("name")
         self._id = self._wandb_init.get("id")
         self._checkpoint_name = checkpoint_name
+        self.LOGGER_JOIN_CHAR = join_char or self.LOGGER_JOIN_CHAR
 
     def __getstate__(self) -> dict[str, Any]:
         import wandb
@@ -440,6 +443,8 @@ def log_metrics(self, metrics: Mapping[str, float], step: Optional[int] = None)
         assert rank_zero_only.rank == 0, "experiment tried to log from global_rank != 0"
 
         metrics = _add_prefix(metrics, self._prefix, self.LOGGER_JOIN_CHAR)
+        join_char = getattr(self, "LOGGER_JOIN_CHAR", ".")
+        metrics = {k.replace("/", join_char): v for k, v in metrics.items()}
         if step is not None and not self._wandb_init.get("sync_tensorboard"):
             self.experiment.log(dict(metrics, **{"trainer/global_step": step}))
         else:

tests/tests_pytorch/loggers/test_wandb_logger_join_char.py

@@ -0,0 +1,51 @@
+from lightning.pytorch.loggers import WandbLogger
+
+
+def test_wandb_logger_custom_join_char(wandb_mock):
+    """Verify that WandbLogger correctly formats metric keys and logs them according to the specified join_char
+    value."""
+
+    # Case 1: Default join_char -> metrics should be logged with hyphens
+    wandb_mock.run = None
+    logger = WandbLogger()  # default join_char = "-"
+    logger.log_metrics({"train/loss": 0.5}, step=1)
+
+    print("\nACTUAL:", wandb_mock.init().log.call_args)
+
+    # Expected output: metrics logged using hyphenated keys
+    wandb_mock.init().log.assert_called_with({"train-loss": 0.5, "trainer/global_step": 1})
+
+    # Case 2: Custom join_char="/" -> metrics should keep their original keys
+    wandb_mock.init().log.reset_mock()
+    logger = WandbLogger(join_char="/")
+    logger.log_metrics({"train/loss": 0.7}, step=2)
+
+    # Expected output: metrics logged with unmodified key names
+    wandb_mock.init().log.assert_called_with({"train/loss": 0.7, "trainer/global_step": 2})
+
+    # Case 3: Custom join_char="_" -> metrics should use underscores
+    wandb_mock.init().log.reset_mock()
+    logger = WandbLogger(join_char="_")
+    logger.log_metrics({"train/loss": 0.9}, step=3)
+
+    # Expected output: metrics logged with underscores in key names
+    wandb_mock.init().log.assert_called_with({"train_loss": 0.9, "trainer/global_step": 3})
+
+
+def test_join_char_persists_across_runs(wandb_mock):
+    """Ensure that a user-defined join_char is consistently applied across multiple logs."""
+    wandb_mock.run = None
+    logger = WandbLogger(join_char="-")
+
+    # Log multiple metric sets with the same logger instance
+    logger.log_metrics({"val/loss": 0.1}, step=1)
+    logger.log_metrics({"val/acc": 0.95}, step=2)
+
+    # Retrieve the logged calls for inspection
+    calls = wandb_mock.init().log.call_args_list
+    logged_1 = calls[0].args[0] if calls[0].args else calls[0].kwargs
+    logged_2 = calls[1].args[0] if calls[1].args else calls[1].kwargs
+
+    # Expected output: both logs use the same join_char formatting (hyphens)
+    assert "val-loss" in logged_1
+    assert "val-acc" in logged_2