tiny fix

2024-07-02 23:06:13 +08:00 · 2024-07-02 23:06:13 +08:00 · 71cdf8956e
parent 821bb6660e
commit 71cdf8956e
1 changed files with 1 additions and 1 deletions
--- a/src/llamafactory/train/ppo/trainer.py
+++ b/src/llamafactory/train/ppo/trainer.py
@ -393,7 +393,7 @@ class CustomPPOTrainer(PPOTrainer, Trainer):
        else:
            reward_model = self.reward_model

-        with unwrap_model_for_generation(reward_model, self.accelerator), self.amp_context:  # support bf16
+        with self.amp_context:  # support bf16
            _, _, values = reward_model(**batch, return_dict=True, use_cache=False)

        if self.finetuning_args.reward_model_type == "lora":