fix mrope

Former-commit-id: 55bee1d333549ca19858b3f5c1b7b86926e5fb09

src/llamafactory/train/rm/trainer.py

@@ -25,7 +25,7 @@ from transformers import Trainer
 from typing_extensions import override
 
 from ...extras import logging
-from ...extras.packages import is_transformers_version_equal_to_4_46
+from ...extras.packages import is_transformers_version_equal_to_4_46, is_transformers_version_greater_than
 from ..callbacks import FixValueHeadModelCallback, PissaConvertCallback, SaveProcessorCallback
 from ..trainer_utils import create_custom_optimizer, create_custom_scheduler
 
@@ -48,6 +48,9 @@ class PairwiseTrainer(Trainer):
     def __init__(
         self, finetuning_args: "FinetuningArguments", processor: Optional["ProcessorMixin"], **kwargs
     ) -> None:
+        if is_transformers_version_greater_than("4.46"):
+            kwargs["processing_class"] = kwargs.pop("tokenizer")
+
         super().__init__(**kwargs)
         self.finetuning_args = finetuning_args
         self.can_return_loss = True  # override property to return eval_loss

src/llamafactory/train/rm/workflow.py

@@ -44,7 +44,9 @@ def run_rm(
     template = get_template_and_fix_tokenizer(tokenizer, data_args)
     dataset_module = get_dataset(template, model_args, data_args, training_args, stage="rm", **tokenizer_module)
     model = load_model(tokenizer, model_args, finetuning_args, training_args.do_train, add_valuehead=True)
-    data_collator = PairwiseDataCollatorWithPadding(template=template, pad_to_multiple_of=8, **tokenizer_module)
+    data_collator = PairwiseDataCollatorWithPadding(
+        template=template, model=model, pad_to_multiple_of=8, **tokenizer_module
+    )
 
     # Update arguments
     training_args.remove_unused_columns = False  # important for multimodal and pairwise dataset