support disable shuffling

Former-commit-id: 9d8c35fd6b838ede0bd6827c6c6121f2cba2b11b

src/llamafactory/train/dpo/trainer.py

@@ -19,7 +19,7 @@ import warnings
 from collections import defaultdict
 from contextlib import nullcontext
 from types import MethodType
-from typing import TYPE_CHECKING, Dict, Literal, Optional, Tuple, Union
+from typing import TYPE_CHECKING, Dict, List, Literal, Optional, Tuple, Union
 
 import torch
 import torch.nn.functional as F
@@ -119,6 +119,13 @@ class CustomDPOTrainer(DPOTrainer):
         create_custom_scheduler(self.args, num_training_steps, optimizer)
         return super().create_scheduler(num_training_steps, optimizer)
 
+    @override
+    def _get_train_sampler(self) -> Optional["torch.utils.data.Sampler"]:
+        if self.finetuning_args.disable_shuffling:
+            return torch.utils.data.SequentialSampler(self.train_dataset)
+
+        return super()._get_train_sampler()
+
     @override
     def get_batch_samples(self, epoch_iterator, num_batches):
         r"""
@@ -266,7 +273,9 @@ class CustomDPOTrainer(DPOTrainer):
         return losses.mean(), metrics
 
     @override
-    def compute_loss(self, model, inputs, return_outputs=False, **kwargs):
+    def compute_loss(
+        self, model: "PreTrainedModel", inputs: Dict[str, "torch.Tensor"], return_outputs: bool = False, **kwargs
+    ) -> Union["torch.Tensor", Tuple["torch.Tensor", List["torch.Tensor"]]]:
         r"""
         Fixes the loss value for transformers 4.46.0.
         https://github.com/huggingface/transformers/blob/v4.46.0/src/transformers/trainer.py#L3605