diff --git a/wit/dataset.py b/wit/dataset/dataset.py
similarity index 92%
rename from wit/dataset.py
rename to wit/dataset/dataset.py
index a9effde..3979b06 100644
--- a/wit/dataset.py
+++ b/wit/dataset/dataset.py
@@ -1,5 +1,5 @@
-from meaning_dataset import MeaningDataset, BatchGroupMeaningDataloader
-from special_dataset import SpecialDataset
+from dataset.meaning_dataset import MeaningDataset, BatchGroupMeaningDataloader
+from dataset.special_dataset import SpecialDataset
 from torch.utils.data import random_split, DataLoader
 
 
diff --git a/wit/meaning_dataset.py b/wit/dataset/meaning_dataset.py
similarity index 100%
rename from wit/meaning_dataset.py
rename to wit/dataset/meaning_dataset.py
diff --git a/wit/special_dataset.py b/wit/dataset/special_dataset.py
similarity index 100%
rename from wit/special_dataset.py
rename to wit/dataset/special_dataset.py
diff --git a/wit/train.py b/wit/train.py
index 88a157d..f8b6539 100644
--- a/wit/train.py
+++ b/wit/train.py
@@ -6,7 +6,7 @@ from wit.model.tokenization_qwen import QWenTokenizer
 from logger import MLFLogger
 
 import configuration
-import dataset as ds
+import dataset.dataset as ds
 
 if __name__ == "__main__":