meilleur gestion de la reprise du trainning

Finetunning/finetunning.py

 # [7/7] Training
 # ----------------------------
 print(f"{80 * '_'}\n[7/7] Starting training...")
-try:
+checkpoint_exists = any(
+    d.startswith("checkpoint-")
+    for d in os.listdir(OUTPUT_DIR)
+) if os.path.exists(OUTPUT_DIR) else False
+
+if checkpoint_exists:
+    print("Checkpoint found → resuming training")
     train_output = trainer.train(resume_from_checkpoint=True)
-except Exception as e:
-    print("No checkpoint found or resume failed, starting fresh training.")
-    print(f"Reason: {e}")
+else:
+    print("No checkpoint found → starting fresh training")
     train_output = trainer.train()
 
+
 print("\n=== Training summary ===")
 print(f"Global steps: {train_output.global_step}")
 print(f"Training loss: {train_output.training_loss}")