diff --git a/deepmd/pt/train/training.py b/deepmd/pt/train/training.py
index 3b8b5a435c..2aa672bd60 100644
--- a/deepmd/pt/train/training.py
+++ b/deepmd/pt/train/training.py
@@ -31,7 +31,6 @@
     TensorLoss,
 )
 from deepmd.pt.model.model import (
-    DOSModel,
     get_model,
     get_zbl_model,
 )
@@ -601,15 +600,13 @@ def single_model_finetune(
                         _finetune_rule_single,
                         _sample_func,
                     ):
-                        # need fix for DOSModel
-                        if not isinstance(_model, DOSModel):
-                            _model = _model_change_out_bias(
-                                _model,
-                                _sample_func,
-                                _bias_adjust_mode="change-by-statistic"
-                                if not _finetune_rule_single.get_random_fitting()
-                                else "set-by-statistic",
-                            )
+                        _model = _model_change_out_bias(
+                            _model,
+                            _sample_func,
+                            _bias_adjust_mode="change-by-statistic"
+                            if not _finetune_rule_single.get_random_fitting()
+                            else "set-by-statistic",
+                        )
                         return _model
 
                     if not self.multi_task:
diff --git a/deepmd/pt/utils/stat.py b/deepmd/pt/utils/stat.py
index b96310a2e6..b0bbda5dbe 100644
--- a/deepmd/pt/utils/stat.py
+++ b/deepmd/pt/utils/stat.py
@@ -399,6 +399,10 @@ def compute_output_stats_global(
     model_pred: Optional[Dict[str, np.ndarray]] = None,
 ):
     """This function only handle stat computation from reduced global labels."""
+    # return directly if model predict is empty for global
+    if model_pred == {}:
+        return {}, {}
+
     # get label dict from sample; for each key, only picking the system with global labels.
     outputs = {
         kk: [
diff --git a/source/tests/pt/test_finetune.py b/source/tests/pt/test_finetune.py
index 2db3076da2..f92dde2cff 100644
--- a/source/tests/pt/test_finetune.py
+++ b/source/tests/pt/test_finetune.py
@@ -44,6 +44,7 @@
 )
 
 from .model.test_permutation import (
+    model_dos,
     model_dpa1,
     model_dpa2,
     model_se_e2_a,
@@ -72,6 +73,13 @@
         must=False,
         high_prec=False,
     ),
+    DataRequirementItem(
+        "dos",
+        ndof=250,
+        atomic=False,
+        must=False,
+        high_prec=True,
+    ),
     DataRequirementItem(
         "atom_ener",
         ndof=1,
@@ -92,6 +100,7 @@
 
 class FinetuneTest:
     def test_finetune_change_out_bias(self):
+        self.testkey = "energy" if self.testkey is None else self.testkey
         # get data
         data = DpLoaderSet(
             self.data_file,
@@ -108,7 +117,7 @@ def test_finetune_change_out_bias(self):
         model = get_model(self.config["model"]).to(env.DEVICE)
         atomic_model = model.atomic_model
         atomic_model["out_bias"] = torch.rand_like(atomic_model["out_bias"])
-        energy_bias_before = to_numpy_array(atomic_model["out_bias"])[0].ravel()
+        energy_bias_before = to_numpy_array(atomic_model["out_bias"])[0]
 
         # prepare original model for test
         dp = torch.jit.script(model)
@@ -123,7 +132,7 @@ def test_finetune_change_out_bias(self):
             sampled,
             bias_adjust_mode="change-by-statistic",
         )
-        energy_bias_after = to_numpy_array(atomic_model["out_bias"])[0].ravel()
+        energy_bias_after = to_numpy_array(atomic_model["out_bias"])[0]
 
         # get ground-truth energy bias change
         sorter = np.argsort(full_type_map)
@@ -140,10 +149,10 @@ def test_finetune_change_out_bias(self):
             to_numpy_array(sampled[0]["box"][:ntest]),
             to_numpy_array(sampled[0]["atype"][0]),
         )[0]
-        energy_diff = to_numpy_array(sampled[0]["energy"][:ntest]) - energy
+        energy_diff = to_numpy_array(sampled[0][self.testkey][:ntest]) - energy
         finetune_shift = (
             energy_bias_after[idx_type_map] - energy_bias_before[idx_type_map]
-        )
+        ).ravel()
         ground_truth_shift = np.linalg.lstsq(atom_nums, energy_diff, rcond=None)[
             0
         ].reshape(-1)
@@ -262,6 +271,7 @@ def setUp(self):
         self.config["training"]["numb_steps"] = 1
         self.config["training"]["save_freq"] = 1
         self.mixed_types = False
+        self.testkey = None
 
 
 class TestEnergyZBLModelSeA(FinetuneTest, unittest.TestCase):
@@ -276,6 +286,22 @@ def setUp(self):
         self.config["training"]["numb_steps"] = 1
         self.config["training"]["save_freq"] = 1
         self.mixed_types = False
+        self.testkey = None
+
+
+class TestEnergyDOSModelSeA(FinetuneTest, unittest.TestCase):
+    def setUp(self):
+        input_json = str(Path(__file__).parent / "dos/input.json")
+        with open(input_json) as f:
+            self.config = json.load(f)
+        self.data_file = [str(Path(__file__).parent / "dos/data/global_system")]
+        self.config["training"]["training_data"]["systems"] = self.data_file
+        self.config["training"]["validation_data"]["systems"] = self.data_file
+        self.config["model"] = deepcopy(model_dos)
+        self.config["training"]["numb_steps"] = 1
+        self.config["training"]["save_freq"] = 1
+        self.mixed_types = False
+        self.testkey = "dos"
 
 
 class TestEnergyModelDPA1(FinetuneTest, unittest.TestCase):
@@ -290,6 +316,7 @@ def setUp(self):
         self.config["training"]["numb_steps"] = 1
         self.config["training"]["save_freq"] = 1
         self.mixed_types = True
+        self.testkey = None
 
 
 class TestEnergyModelDPA2(FinetuneTest, unittest.TestCase):
@@ -306,6 +333,7 @@ def setUp(self):
         self.config["training"]["numb_steps"] = 1
         self.config["training"]["save_freq"] = 1
         self.mixed_types = True
+        self.testkey = None
 
 
 if __name__ == "__main__":