From 3b73dbc0b05162180b3bb737c26fda555c9a398f Mon Sep 17 00:00:00 2001
From: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Date: Tue, 14 Jul 2026 22:14:43 +0000
Subject: [PATCH] fix(utils): stop clobbering litellm.num_retries global after
 a completion exception

---
 litellm/utils.py                 | 28 ++--------------------------
 tests/test_litellm/test_utils.py | 23 +++++++++++++++++++++++
 2 files changed, 25 insertions(+), 26 deletions(-)

diff --git a/litellm/utils.py b/litellm/utils.py
index 18b89ee0d13..1a98b6ef96f 100644
--- a/litellm/utils.py
+++ b/litellm/utils.py
@@ -1501,18 +1501,7 @@ def client(original_function):
         except Exception as e:
             call_type = original_function.__name__
             if call_type == CallTypes.completion.value:
-                num_retries = kwargs.get("num_retries", None) or litellm.num_retries or None
-                if kwargs.get("retry_policy", None):
-                    get_num_retries_from_retry_policy = getattr(
-                        sys.modules[__name__], "get_num_retries_from_retry_policy"
-                    )
-                    reset_retry_policy = getattr(sys.modules[__name__], "reset_retry_policy")
-                    num_retries = get_num_retries_from_retry_policy(
-                        exception=e,
-                        retry_policy=kwargs.get("retry_policy"),
-                    )
-                    kwargs["retry_policy"] = reset_retry_policy()  # prevent infinite loops
-                litellm.num_retries = None  # set retries to None to prevent infinite loops
+                num_retries, kwargs = _get_wrapper_num_retries(kwargs=kwargs, exception=e)
                 context_window_fallback_dict = kwargs.get("context_window_fallback_dict", {})
 
                 _is_litellm_router_call = "model_group" in (
@@ -1540,18 +1529,7 @@ def client(original_function):
                         kwargs["model"] = context_window_fallback_dict[model]
                     return original_function(*args, **kwargs)
             elif call_type == CallTypes.responses.value:
-                num_retries = kwargs.get("num_retries", None) or litellm.num_retries or None
-                if kwargs.get("retry_policy", None):
-                    get_num_retries_from_retry_policy = getattr(
-                        sys.modules[__name__], "get_num_retries_from_retry_policy"
-                    )
-                    reset_retry_policy = getattr(sys.modules[__name__], "reset_retry_policy")
-                    num_retries = get_num_retries_from_retry_policy(
-                        exception=e,
-                        retry_policy=kwargs.get("retry_policy"),
-                    )
-                    kwargs["retry_policy"] = reset_retry_policy()  # prevent infinite loops
-                litellm.num_retries = None  # set retries to None to prevent infinite loops
+                num_retries, kwargs = _get_wrapper_num_retries(kwargs=kwargs, exception=e)
 
                 _is_litellm_router_call = "model_group" in (
                     kwargs.get("metadata") or {}
@@ -1822,7 +1800,6 @@ def client(original_function):
                     num_retries and not _is_litellm_router_call
                 ):  # only enter this if call is not from litellm router/proxy. router has it's own logic for retrying
                     try:
-                        litellm.num_retries = None  # set retries to None to prevent infinite loops
                         kwargs["num_retries"] = num_retries
                         kwargs["original_function"] = original_function
                         if isinstance(e, openai.RateLimitError):  # rate limiting specific error
@@ -1852,7 +1829,6 @@ def client(original_function):
                     num_retries and not _is_litellm_router_call
                 ):  # only enter this if call is not from litellm router/proxy. router has it's own logic for retrying
                     try:
-                        litellm.num_retries = None  # set retries to None to prevent infinite loops
                         kwargs["num_retries"] = num_retries
                         kwargs["original_function"] = original_function
                         if isinstance(e, openai.RateLimitError):  # rate limiting specific error
diff --git a/tests/test_litellm/test_utils.py b/tests/test_litellm/test_utils.py
index 1a5c419754b..6432eade0ef 100644
--- a/tests/test_litellm/test_utils.py
+++ b/tests/test_litellm/test_utils.py
@@ -4737,3 +4737,26 @@ def test_gemini_image_models_do_not_support_reasoning(
         f"{model} incorrectly classified as reasoning-capable. "
         "Add 'supports_reasoning: false' to its model_cost entry."
     )
+
+
+def test_completion_exception_preserves_global_num_retries(monkeypatch):
+    monkeypatch.setattr(litellm, "num_retries", 2)
+    with pytest.raises(litellm.RateLimitError):
+        litellm.completion(
+            model="gpt-4o-mini",
+            messages=[{"role": "user", "content": "hi"}],
+            mock_response="litellm.RateLimitError",
+        )
+    assert litellm.num_retries == 2
+
+
+@pytest.mark.asyncio
+async def test_acompletion_exception_preserves_global_num_retries(monkeypatch):
+    monkeypatch.setattr(litellm, "num_retries", 2)
+    with pytest.raises(litellm.RateLimitError):
+        await litellm.acompletion(
+            model="gpt-4o-mini",
+            messages=[{"role": "user", "content": "hi"}],
+            mock_response="litellm.RateLimitError",
+        )
+    assert litellm.num_retries == 2