use accelerate to determin logging

src/lerobot/scripts/lerobot_train.py

@@ -151,6 +151,8 @@ def train(cfg: TrainPipelineConfig, accelerator: Accelerator | None = None):
 
         ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters=True)
         accelerator = Accelerator(step_scheduler_with_optimizer=False, kwargs_handlers=[ddp_kwargs])
+    
+    init_logging(accelerator=accelerator)
 
     # Determine if this is the main process (for logging and checkpointing)
     # When using accelerate, only the main process should log to avoid duplicate outputs
@@ -435,7 +437,6 @@ def train(cfg: TrainPipelineConfig, accelerator: Accelerator | None = None):
 
 
 def main():
-    init_logging()
     train()
 
 

src/lerobot/utils/utils.py

@@ -143,9 +143,8 @@ def init_logging(
     logger.handlers.clear()
 
     # Determine if this is a non-main process in distributed training
-    # Check LOCAL_RANK env var (set by accelerate/torchrun)
-    local_rank = int(os.environ.get("LOCAL_RANK", -1))
-    is_main_process = local_rank <= 0  # -1 means not distributed, 0 means main process
+    if accelerator is not None:
+        is_main_process = accelerator.is_main_process
 
     # Console logging (main process only)
     if is_main_process: