diff --git a/litellm/caching.py b/litellm/caching.py
index 5601fab2fd..59fc0ab672 100644
--- a/litellm/caching.py
+++ b/litellm/caching.py
@@ -678,7 +678,7 @@ class Cache:
                         if k == "ttl":
                             kwargs["ttl"] = v
                 cached_data = {"timestamp": time.time(), "response": result}
-                return cache_key, cached_data
+                return cache_key, cached_data, kwargs
             else:
                 raise Exception("cache key is None")
         except Exception as e:
@@ -696,7 +696,7 @@ class Cache:
             None
         """
         try:
-            cache_key, cached_data = self._add_cache_logic(
+            cache_key, cached_data, kwargs = self._add_cache_logic(
                 result=result, *args, **kwargs
             )
             self.cache.set_cache(cache_key, cached_data, **kwargs)
@@ -710,7 +710,7 @@ class Cache:
         Async implementation of add_cache
         """
         try:
-            cache_key, cached_data = self._add_cache_logic(
+            cache_key, cached_data, kwargs = self._add_cache_logic(
                 result=result, *args, **kwargs
             )
             await self.cache.async_set_cache(cache_key, cached_data, **kwargs)
diff --git a/litellm/tests/test_caching.py b/litellm/tests/test_caching.py
index 3250a26215..695ad931a2 100644
--- a/litellm/tests/test_caching.py
+++ b/litellm/tests/test_caching.py
@@ -109,10 +109,7 @@ def test_caching_with_cache_controls():
         )
         print(f"response1: {response1}")
         print(f"response2: {response2}")
-        assert (
-            response2["choices"][0]["message"]["content"]
-            == response1["choices"][0]["message"]["content"]
-        )
+        assert response2["id"] == response1["id"]
     except Exception as e:
         print(f"error occurred: {traceback.format_exc()}")
         pytest.fail(f"Error occurred: {e}")
diff --git a/litellm/utils.py b/litellm/utils.py
index 6059cdf85a..3fee13937a 100644
--- a/litellm/utils.py
+++ b/litellm/utils.py
@@ -2029,7 +2029,11 @@ def client(original_function):
             # if caching is false or cache["no-cache"]==True, don't run this
             if (
                 (
-                    (kwargs.get("caching", None) is None and litellm.cache is not None)
+                    (
+                        kwargs.get("caching", None) is None
+                        and kwargs.get("cache", None) is None
+                        and litellm.cache is not None
+                    )
                     or kwargs.get("caching", False) == True
                     or (
                         kwargs.get("cache", None) is not None