fix

src/lightning/fabric/utilities/distributed.py

@@ -297,10 +297,11 @@ def _init_dist_connection(
         # PyTorch >= 2.4 warns about undestroyed NCCL process group, so we need to do it at program exit
         atexit.register(_destroy_dist_connection)
 
-    # On rank=0 let everyone know training is starting
-    rank_zero_info(
+    # On local_rank=0 let everyone know training is starting
+    if cluster_environment.local_rank() == 0:
+        log.info(
             f"{'-' * 100}\n"
-        f"distributed_backend={torch_distributed_backend}\n"
+            f"Distributed backend: {torch_distributed_backend.upper()}\n"
             f"All distributed processes registered. Starting with {world_size} processes\n"
             f"{'-' * 100}\n"
         )

src/lightning/pytorch/callbacks/progress/progress_bar.py

@@ -11,6 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import logging
 from typing import Any, Dict, Optional, Union
 
 from typing_extensions import override
@@ -19,6 +20,8 @@ import lightning.pytorch as pl
 from lightning.pytorch.callbacks import Callback
 from lightning.pytorch.utilities.rank_zero import rank_zero_warn
 
+_log = logging.getLogger(__name__)
+
 
 class ProgressBar(Callback):
     r"""The base class for progress bars in Lightning. It is a :class:`~lightning.pytorch.callbacks.Callback` that keeps
@@ -173,6 +176,8 @@ class ProgressBar(Callback):
         self._trainer = trainer
         if not trainer.is_global_zero:
             self.disable()
+        if trainer.node_rank > 0 and trainer.local_rank == 0:
+            _log.info("The progress bar output will appear on node 0.")
 
     def get_metrics(
         self, trainer: "pl.Trainer", pl_module: "pl.LightningModule"

src/lightning/pytorch/callbacks/progress/rich_progress.py

@@ -11,7 +11,6 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-import logging
 import math
 from dataclasses import dataclass
 from datetime import timedelta
@@ -25,7 +24,6 @@ from lightning.pytorch.callbacks.progress.progress_bar import ProgressBar
 from lightning.pytorch.utilities.types import STEP_OUTPUT
 
 _RICH_AVAILABLE = RequirementCache("rich>=10.2.2")
-_log = logging.getLogger(__name__)
 
 
 if _RICH_AVAILABLE:
@@ -366,10 +364,6 @@ class RichProgressBar(ProgressBar):
         if self.progress:
             self.progress.refresh()
 
-    def setup(self, trainer: "pl.Trainer", pl_module: "pl.LightningModule", stage: str) -> None:
-        if trainer.node_rank > 0 and trainer.local_rank == 0:
-            _log.info("The progress bar output will appear on node 0.")
-
     @override
     def on_train_start(self, trainer: "pl.Trainer", pl_module: "pl.LightningModule") -> None:
         self._init_progress(trainer)

src/lightning/pytorch/callbacks/progress/tqdm_progress.py

@@ -12,7 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import importlib
-import logging
 import math
 import os
 import sys
@@ -34,7 +33,6 @@ import lightning.pytorch as pl
 from lightning.pytorch.callbacks.progress.progress_bar import ProgressBar
 from lightning.pytorch.utilities.rank_zero import rank_zero_debug
 
-_log = logging.getLogger(__name__)
 _PAD_SIZE = 5
 
 
@@ -248,10 +246,6 @@ class TQDMProgressBar(ProgressBar):
             bar_format=self.BAR_FORMAT,
         )
 
-    def setup(self, trainer: "pl.Trainer", pl_module: "pl.LightningModule", stage: str) -> None:
-        if trainer.node_rank > 0 and trainer.local_rank == 0:
-            _log.info("The progress bar output will appear on node 0.")
-
     @override
     def on_sanity_check_start(self, *_: Any) -> None:
         self.val_progress_bar = self.init_sanity_tqdm()