diff --git a/narwhals/_arrow/utils.py b/narwhals/_arrow/utils.py
index 0012860d1..a821677f5 100644
--- a/narwhals/_arrow/utils.py
+++ b/narwhals/_arrow/utils.py
@@ -6,6 +6,7 @@
 from typing import Generator
 from typing import Sequence
 
+from narwhals.translate import to_py_scalar
 from narwhals.utils import isinstance_or_issubclass
 
 if TYPE_CHECKING:
@@ -463,6 +464,6 @@ def _rolling(
 
         if num_valid >= min_periods:
             valid_weights = weights_.slice(0, num_valid)
-            yield aggregate_function(valid_window, valid_weights)
+            yield to_py_scalar(aggregate_function(valid_window, valid_weights))
         else:
             yield None
diff --git a/narwhals/_pandas_like/series.py b/narwhals/_pandas_like/series.py
index 82cf2c4f6..1150ec908 100644
--- a/narwhals/_pandas_like/series.py
+++ b/narwhals/_pandas_like/series.py
@@ -694,7 +694,7 @@ def rolling_mean(
         if weights is not None:
             msg = (
                 "`weights` argument is not supported in `rolling_mean` for "
-                f"{self._implementation} backend."
+                "pandas-like backend."
             )
             raise NotImplementedError(msg)
 
diff --git a/tests/expr_and_series/rolling_mean_test.py b/tests/expr_and_series/rolling_mean_test.py
index d99086f25..ad318e50d 100644
--- a/tests/expr_and_series/rolling_mean_test.py
+++ b/tests/expr_and_series/rolling_mean_test.py
@@ -63,7 +63,7 @@ def test_weighted_rolling_mean_expr(constructor: Constructor) -> None:
             NotImplementedError,
             match=re.escape("`weights` argument is not supported in `rolling_mean`"),
         )
-        if "pandas" in str(constructor) or "dask" in str(constructor)
+        if any(x in str(constructor) for x in ("pandas", "modin", "cudf", "dask"))
         else does_not_raise()
     )
     df = nw.from_native(constructor(data_weighted))
@@ -82,7 +82,7 @@ def test_weighted_rolling_mean_series(constructor_eager: ConstructorEager) -> No
             NotImplementedError,
             match=re.escape("`weights` argument is not supported in `rolling_mean`"),
         )
-        if "pandas" in str(constructor_eager)
+        if any(x in str(constructor_eager) for x in ("pandas", "modin", "cudf"))
         else does_not_raise()
     )
     df = nw.from_native(constructor_eager(data_weighted), eager_only=True)