From 51e114b1ee91f88067c23c6343b0c238eede8a90 Mon Sep 17 00:00:00 2001 From: Alex Date: Thu, 15 Jan 2026 17:03:18 +0100 Subject: [PATCH] meilleur gestion de la reprise du trainning --- Finetunning/finetunning.py | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-) diff --git a/Finetunning/finetunning.py b/Finetunning/finetunning.py index 9970578..a2142f0 100644 --- a/Finetunning/finetunning.py +++ b/Finetunning/finetunning.py @@ -182,13 +182,19 @@ print("Trainer initialized.") # [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}")