diff --git a/nvflare/apis/analytix.py b/nvflare/apis/analytix.py
index b311ab19d2..b74841d969 100644
--- a/nvflare/apis/analytix.py
+++ b/nvflare/apis/analytix.py
@@ -93,7 +93,7 @@ def __init__(
             sender (LogWriterName): Type of sender for syntax such as Tensorboard or MLflow
             kwargs (optional, dict): additional arguments to be passed.
         """
-        self._validate_data_types(data_type, key, value, **kwargs)
+        value = self._validate_data_types(data_type, key, value, **kwargs)
         self.tag = key
         self.value = value
         self.data_type = data_type
@@ -183,22 +183,56 @@ def _validate_data_types(
         path = kwargs.get(TrackConst.PATH_KEY, None)
         if path and not isinstance(path, str):
             raise TypeError("expect path to be an instance of str, but got {}.".format(type(step)))
-        if data_type in [AnalyticsDataType.SCALAR, AnalyticsDataType.METRIC] and not (
-            isinstance(value, float) or isinstance(value, int)
-        ):
-            raise TypeError(f"expect '{key}' value to be an instance of float or int, but got '{type(value)}'.")
+        if data_type in [AnalyticsDataType.SCALAR, AnalyticsDataType.METRIC]:
+            is_numeric_scalar, normalized_value = self._normalize_numeric_scalar(value)
+            if not is_numeric_scalar:
+                raise TypeError(f"expect '{key}' value to be an instance of float or int, but got '{type(value)}'.")
+            value = normalized_value
         elif data_type in [
             AnalyticsDataType.METRICS,
             AnalyticsDataType.PARAMETERS,
             AnalyticsDataType.SCALARS,
         ] and not isinstance(value, dict):
             raise TypeError(f"expect '{key}' value to be an instance of dict, but got '{type(value)}'.")
+        elif data_type in [AnalyticsDataType.METRICS, AnalyticsDataType.SCALARS]:
+            normalized_dict = {}
+            for k, v in value.items():
+                is_numeric_scalar, normalized_value = self._normalize_numeric_scalar(v)
+                normalized_dict[k] = normalized_value if is_numeric_scalar else v
+            value = normalized_dict
         elif data_type == AnalyticsDataType.TEXT and not isinstance(value, str):
             raise TypeError(f"expect '{key}' value to be an instance of str, but got '{type(value)}'.")
         elif data_type == AnalyticsDataType.TAGS and not isinstance(value, dict):
             raise TypeError(
                 f"expect '{key}' data type expects value to be an instance of dict, but got '{type(value)}'"
             )
+        return value
+
+    @staticmethod
+    def _normalize_numeric_scalar(value):
+        if isinstance(value, (float, int)):
+            return True, value
+
+        item = getattr(value, "item", None)
+        if not callable(item):
+            return False, value
+
+        shape = getattr(value, "shape", None)
+        if shape is not None:
+            try:
+                if tuple(shape) != ():
+                    return False, value
+            except TypeError:
+                return False, value
+
+        try:
+            scalar = item()
+        except (TypeError, ValueError):
+            return False, value
+
+        if isinstance(scalar, (float, int)):
+            return True, scalar
+        return False, value
 
     @classmethod
     def convert_data_type(
diff --git a/tests/unit_test/apis/analytix_test.py b/tests/unit_test/apis/analytix_test.py
index ad068c704c..55fbc469d3 100644
--- a/tests/unit_test/apis/analytix_test.py
+++ b/tests/unit_test/apis/analytix_test.py
@@ -105,3 +105,53 @@ def test_to_dxo(self, data: AnalyticsData):
     def test_from_dxo_invalid(self, dxo, expected_error, expected_msg):
         with pytest.raises(expected_error, match=expected_msg):
             _ = AnalyticsData.from_dxo(dxo)
+
+    def test_scalar_like_value_is_normalized(self):
+        class ScalarLike:
+            shape = ()
+
+            def item(self):
+                return 1.25
+
+        data = AnalyticsData(key="loss", value=ScalarLike(), data_type=AnalyticsDataType.SCALAR)
+
+        assert data.value == 1.25
+        assert isinstance(data.value, float)
+
+    def test_numeric_dict_values_are_normalized(self):
+        class ScalarLike:
+            shape = ()
+
+            def item(self):
+                return 1.25
+
+        data = AnalyticsData(
+            key="losses",
+            value={"train": ScalarLike(), "valid": 2},
+            data_type=AnalyticsDataType.SCALARS,
+        )
+
+        assert data.value == {"train": 1.25, "valid": 2}
+        assert isinstance(data.value["train"], float)
+
+    def test_numpy_numeric_values_are_normalized(self):
+        np = pytest.importorskip("numpy")
+
+        data = AnalyticsData(key="loss", value=np.float32(1.25), data_type=AnalyticsDataType.SCALAR)
+        dxo = create_analytic_dxo(
+            tag="loss", value=np.asarray(1.25, dtype=np.float32), data_type=AnalyticsDataType.SCALAR
+        )
+        metrics = AnalyticsData(
+            key="losses",
+            value={"train": np.float32(1.25), "valid": np.asarray(2, dtype=np.int32)},
+            data_type=AnalyticsDataType.SCALARS,
+        )
+
+        assert data.value == pytest.approx(1.25)
+        assert isinstance(data.value, float)
+        assert dxo.data[TrackConst.TRACK_VALUE] == pytest.approx(1.25)
+        assert isinstance(dxo.data[TrackConst.TRACK_VALUE], float)
+        assert metrics.value["train"] == pytest.approx(1.25)
+        assert metrics.value["valid"] == 2
+        assert isinstance(metrics.value["train"], float)
+        assert isinstance(metrics.value["valid"], int)