diff --git a/pts/dataset/repository/_m5.py b/pts/dataset/repository/_m5.py
index cd70b0b..bc26599 100644
--- a/pts/dataset/repository/_m5.py
+++ b/pts/dataset/repository/_m5.py
@@ -7,15 +7,21 @@ import numpy as np
 import pandas as pd
 
 from gluonts.dataset.field_names import FieldName
-from gluonts.dataset.repository._util import metadata, save_to_file
+from gluonts.dataset.repository._util import metadata
 from gluonts.time_feature.holiday import squared_exponential_kernel
 from pts.feature import CustomDateFeatureSet
 
+from gluonts.dataset import DatasetWriter
+from gluonts.dataset.common import MetaData, TrainDatasets
+
+from pathlib import Path
 
 def generate_pts_m5_dataset(
     dataset_path: Path,
     pandas_freq: str,
-    prediction_length: int = 28,
+    prediction_length: int,
+    m5_file_path: Path,
+    dataset_writer: DatasetWriter,
     alpha: float = 0.5,
 ):
     cal_path = f"{dataset_path}/calendar.csv"
@@ -134,7 +140,6 @@ def generate_pts_m5_dataset(
     ]
 
     # Build training set
-    train_file = dataset_path / "train" / "data.json"
     train_ds = []
     for index, item in sales_train_validation.iterrows():
         id, item_id, dept_id, cat_id, store_id, state_id = index
@@ -178,30 +183,17 @@ def generate_pts_m5_dataset(
             .astype(np.float32)
             .tolist()
         )
+        d = {
+            FieldName.TARGET: time_series["target"],
+            FieldName.START: time_series["start"],
+            FieldName.FEAT_DYNAMIC_REAL: time_series["feat_dynamic_real"],
+            FieldName.FEAT_STATIC_CAT: time_series["feat_static_cat"],
+            FieldName.ITEM_ID: time_series["item_id"],
+        }
 
-        train_ds.append(time_series.copy())
-
-    # Build training set
-    train_file = dataset_path / "train" / "data.json"
-    save_to_file(train_file, train_ds)
-
-    # Create metadata file
-    meta_file = dataset_path / "metadata.json"
-    with open(meta_file, "w") as f:
-        f.write(
-            json.dumps(
-                {
-                    "freq": pandas_freq,
-                    "prediction_length": prediction_length,
-                    "feat_static_cat": feat_static_cat,
-                    "feat_dynamic_real": feat_dynamic_real,
-                    "cardinality": len(train_ds),
-                }
-            )
-        )
+        train_ds.append(d)
 
     # Build testing set
-    test_file = dataset_path / "test" / "data.json"
     test_ds = []
     for index, item in sales_train_evaluation.iterrows():
         id, item_id, dept_id, cat_id, store_id, state_id = index
@@ -245,7 +237,27 @@ def generate_pts_m5_dataset(
             .astype(np.float32)
             .tolist()
         )
+        d = {
+            FieldName.TARGET: time_series["target"],
+            FieldName.START: time_series["start"],
+            FieldName.FEAT_DYNAMIC_REAL: time_series["feat_dynamic_real"],
+            FieldName.FEAT_STATIC_CAT: time_series["feat_static_cat"],
+            FieldName.ITEM_ID: time_series["item_id"],
+        }
 
-        test_ds.append(time_series.copy())
+        test_ds.append(d)
 
-    save_to_file(test_file, test_ds)
+    # Create metadata file
+    meta = MetaData(
+        **metadata(
+            cardinality=len(train_ds),
+            freq=pandas_freq,
+            feat_static_cat=feat_static_cat,
+            feat_dynamic_real=feat_dynamic_real,
+            prediction_length=prediction_length,
+        )
+    )
+    dataset = TrainDatasets(metadata=meta, train=train_ds, test=test_ds)
+    dataset.save(
+        path_str=str(dataset_path), writer=dataset_writer, overwrite=True
+    )