diff --git a/src/lighteval/utils/utils.py b/src/lighteval/utils/utils.py
index 3ab5976d8..7f7006b63 100644
--- a/src/lighteval/utils/utils.py
+++ b/src/lighteval/utils/utils.py
@@ -56,7 +56,7 @@ def rec(nest: dict, prefix: str, into: dict):
                         )  # Need this for markdown
                         into[prefix + k + sep + str(i)] = vv.tolist() if isinstance(vv, np.ndarray) else vv
             elif isinstance(v, np.ndarray):
-                into[prefix + k + sep + str(i)] = v.tolist()
+                into[prefix + k] = v.tolist()
             else:
                 v = clean_markdown(v)
                 into[prefix + k] = v
diff --git a/tests/unit/utils/test_utils.py b/tests/unit/utils/test_utils.py
index 68e844712..7fef084cc 100644
--- a/tests/unit/utils/test_utils.py
+++ b/tests/unit/utils/test_utils.py
@@ -22,7 +22,21 @@
 
 import unittest
 
-from lighteval.utils.utils import remove_reasoning_tags
+import numpy as np
+
+from lighteval.utils.utils import flatten_dict, remove_reasoning_tags
+
+
+class TestFlattenDict(unittest.TestCase):
+    def test_numpy_array_value(self):
+        result = flatten_dict({"scores": np.array([1, 2])})
+
+        self.assertEqual(result, {"scores": [1, 2]})
+
+    def test_numpy_array_value_does_not_reuse_list_index(self):
+        result = flatten_dict({"items": ["a", "b"], "scores": np.array([1, 2])})
+
+        self.assertEqual(result, {"items/0": "a", "items/1": "b", "scores": [1, 2]})
 
 
 class TestRemoveReasoningTags(unittest.TestCase):