fix precommit and fix tests

src/lerobot/scripts/lerobot_train.py

@@ -98,8 +98,8 @@ def update_policy(
     if grad_clip_norm > 0:
         grad_norm = accelerator.clip_grad_norm_(policy.parameters(), grad_clip_norm)
     else:
-       grad_norm = torch.nn.utils.clip_grad_norm_(
+        grad_norm = torch.nn.utils.clip_grad_norm_(
-            policy.parameters(), float('inf'), error_if_nonfinite=False
+            policy.parameters(), float("inf"), error_if_nonfinite=False
         )
 
     # Optimizer step
@@ -151,7 +151,7 @@ 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)

src/lerobot/utils/utils.py

@@ -20,7 +20,6 @@ import select
 import subprocess
 import sys
 import time
-from collections.abc import Callable
 from copy import copy, deepcopy
 from datetime import datetime
 from pathlib import Path
@@ -143,8 +142,7 @@ def init_logging(
     logger.handlers.clear()
 
     # Determine if this is a non-main process in distributed training
-    if accelerator is not None:
+    is_main_process = accelerator.is_main_process if accelerator is not None else True
-        is_main_process = accelerator.is_main_process
 
     # Console logging (main process only)
     if is_main_process: