add docstrings, refactor logger

Former-commit-id: c34e489d71f8f539028543ccf8ee92cecedd6276

src/llamafactory/train/rm/trainer.py

@@ -22,6 +22,7 @@ from typing import TYPE_CHECKING, Dict, List, Optional, Tuple, Union
 
 import torch
 from transformers import Trainer
+from typing_extensions import override
 
 from ...extras.logging import get_logger
 from ..callbacks import FixValueHeadModelCallback, PissaConvertCallback, SaveProcessorCallback
@@ -63,17 +64,20 @@ class PairwiseTrainer(Trainer):
             self.accelerator.clip_grad_norm_ = MethodType(clip_grad_norm_old_version, self.accelerator)
             self.add_callback(BAdamCallback)
 
+    @override
     def create_optimizer(self) -> "torch.optim.Optimizer":
         if self.optimizer is None:
             self.optimizer = create_custom_optimizer(self.model, self.args, self.finetuning_args)
         return super().create_optimizer()
 
+    @override
     def create_scheduler(
         self, num_training_steps: int, optimizer: Optional["torch.optim.Optimizer"] = None
     ) -> "torch.optim.lr_scheduler.LRScheduler":
         create_custom_scheduler(self.args, num_training_steps, optimizer)
         return super().create_scheduler(num_training_steps, optimizer)
 
+    @override
     def compute_loss(
         self, model: "PreTrainedModel", inputs: Dict[str, "torch.Tensor"], return_outputs: bool = False
     ) -> Union["torch.Tensor", Tuple["torch.Tensor", List["torch.Tensor"]]]: