diff --git a/deepmd/dpmodel/common.py b/deepmd/dpmodel/common.py
index d9d57d2d6c..5c75229e49 100644
--- a/deepmd/dpmodel/common.py
+++ b/deepmd/dpmodel/common.py
@@ -50,6 +50,33 @@
 DEFAULT_PRECISION = "float64"
 
 
+def get_xp_precision(
+    xp: Any,
+    precision: str,
+):
+    """Get the precision from the API compatible namespace."""
+    if precision == "float16" or precision == "half":
+        return xp.float16
+    elif precision == "float32" or precision == "single":
+        return xp.float32
+    elif precision == "float64" or precision == "double":
+        return xp.float64
+    elif precision == "int32":
+        return xp.int32
+    elif precision == "int64":
+        return xp.int64
+    elif precision == "bool":
+        return bool
+    elif precision == "default":
+        return get_xp_precision(xp, RESERVED_PRECISON_DICT[PRECISION_DICT[precision]])
+    elif precision == "global":
+        return get_xp_precision(xp, RESERVED_PRECISON_DICT[GLOBAL_NP_FLOAT_PRECISION])
+    elif precision == "bfloat16":
+        return ml_dtypes.bfloat16
+    else:
+        raise ValueError(f"unsupported precision {precision} for {xp}")
+
+
 class NativeOP(ABC):
     """The unit operation of a native model."""
 
diff --git a/deepmd/dpmodel/fitting/general_fitting.py b/deepmd/dpmodel/fitting/general_fitting.py
index 8e9e228787..016f947206 100644
--- a/deepmd/dpmodel/fitting/general_fitting.py
+++ b/deepmd/dpmodel/fitting/general_fitting.py
@@ -18,6 +18,7 @@
     NativeOP,
 )
 from deepmd.dpmodel.common import (
+    get_xp_precision,
     to_numpy_array,
 )
 from deepmd.dpmodel.utils import (
@@ -417,7 +418,9 @@ def _call_common(
 
         # calcualte the prediction
         if not self.mixed_types:
-            outs = xp.zeros([nf, nloc, net_dim_out], dtype=self.prec)
+            outs = xp.zeros(
+                [nf, nloc, net_dim_out], dtype=get_xp_precision(xp, self.precision)
+            )
             for type_i in range(self.ntypes):
                 mask = xp.tile(
                     xp.reshape((atype == type_i), [nf, nloc, 1]), (1, 1, net_dim_out)
@@ -443,4 +446,4 @@ def _call_common(
         exclude_mask = self.emask.build_type_exclude_mask(atype)
         # nf x nloc x nod
         outs = outs * xp.astype(exclude_mask[:, :, None], outs.dtype)
-        return {self.var_name: outs.astype(GLOBAL_NP_FLOAT_PRECISION)}
+        return {self.var_name: xp.astype(outs, get_xp_precision(xp, "global"))}
diff --git a/source/tests/common/test_common.py b/source/tests/common/test_common.py
new file mode 100644
index 0000000000..478c512fed
--- /dev/null
+++ b/source/tests/common/test_common.py
@@ -0,0 +1,27 @@
+# SPDX-License-Identifier: LGPL-3.0-or-later
+import unittest
+
+import array_api_compat
+import ml_dtypes
+import numpy as np
+
+from deepmd.dpmodel.common import (
+    get_xp_precision,
+)
+from deepmd.env import (
+    GLOBAL_NP_FLOAT_PRECISION,
+)
+
+
+class TestGetXPPrecision(unittest.TestCase):
+    def test(self):
+        aa = np.zeros(3)
+        xp = array_api_compat.array_namespace(aa)
+        self.assertTrue(get_xp_precision(xp, "float16"), xp.float16)
+        self.assertTrue(get_xp_precision(xp, "float32"), xp.float32)
+        self.assertTrue(get_xp_precision(xp, "float64"), xp.float64)
+        self.assertTrue(get_xp_precision(xp, "single"), xp.float32)
+        self.assertTrue(get_xp_precision(xp, "double"), xp.float64)
+        self.assertTrue(get_xp_precision(xp, "global"), GLOBAL_NP_FLOAT_PRECISION)
+        self.assertTrue(get_xp_precision(xp, "default"), GLOBAL_NP_FLOAT_PRECISION)
+        self.assertTrue(get_xp_precision(xp, "bfloat16"), ml_dtypes.bfloat16)