diff --git a/lib/trade_models/quant_transformer.py b/lib/trade_models/quant_transformer.py
index 9962896..55a92ec 100755
--- a/lib/trade_models/quant_transformer.py
+++ b/lib/trade_models/quant_transformer.py
@@ -37,7 +37,9 @@ from qlib.data.dataset.handler import DataHandlerLP
 
 default_net_config = dict(d_feat=6, hidden_size=48, depth=5, pos_drop=0.1)
 
-default_opt_config = dict(epochs=200, lr=0.001, batch_size=2000, early_stop=20, loss="mse", optimizer="adam")
+default_opt_config = dict(
+    epochs=200, lr=0.001, batch_size=2000, early_stop=20, loss="mse", optimizer="adam", num_workers=4
+)
 
 
 class QuantTransformer(Model):
@@ -159,6 +161,16 @@ class QuantTransformer(Model):
                 torch.from_numpy(df_data["label"].values).squeeze().float(),
             )
 
+        def _prepare_loader(dataset, shuffle):
+            return th_data.DataLoader(
+                dataset,
+                batch_size=self.opt_config["batch_size"],
+                drop_last=False,
+                pin_memory=True,
+                num_workers=self.opt_config["num_workers"],
+                shuffle=shuffle,
+            )
+
         df_train, df_valid, df_test = dataset.prepare(
             ["train", "valid", "test"],
             col_set=["feature", "label"],
@@ -169,15 +181,10 @@ class QuantTransformer(Model):
             _prepare_dataset(df_valid),
             _prepare_dataset(df_test),
         )
-
-        train_loader = th_data.DataLoader(
-            train_dataset, batch_size=self.opt_config["batch_size"], shuffle=True, drop_last=False, pin_memory=True
-        )
-        valid_loader = th_data.DataLoader(
-            valid_dataset, batch_size=self.opt_config["batch_size"], shuffle=False, drop_last=False, pin_memory=True
-        )
-        test_loader = th_data.DataLoader(
-            test_dataset, batch_size=self.opt_config["batch_size"], shuffle=False, drop_last=False, pin_memory=True
+        train_loader, valid_loader, test_loader = (
+            _prepare_loader(train_dataset, True),
+            _prepare_loader(valid_dataset, False),
+            _prepare_loader(test_dataset, False),
         )
 
         if save_path == None: