fix(enterprise): resolve openai_moderations model at call time and default to omni-moderation-latest

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

enterprise/enterprise_hooks/openai_moderation.py

@@ -17,6 +17,7 @@ from fastapi import HTTPException
 
 import litellm
 from litellm._logging import verbose_proxy_logger
+from litellm.constants import DEFAULT_OPENAI_MODERATIONS_MODEL
 from litellm.integrations.custom_logger import CustomLogger
 from litellm.proxy._types import UserAPIKeyAuth
 from litellm.proxy.guardrails._content_utils import iter_message_text
@@ -24,11 +25,9 @@ from litellm.types.utils import CallTypesLiteral
 
 
 class _ENTERPRISE_OpenAI_Moderation(CustomLogger):
-    def __init__(self):
-        self.model_name = (
-            litellm.openai_moderations_model_name or "text-moderation-latest"
-        )  # pass the model_name you initialized on litellm.Router()
-        pass
+    @property
+    def model_name(self) -> str:
+        return litellm.openai_moderations_model_name or DEFAULT_OPENAI_MODERATIONS_MODEL
 
     #### CALL HOOKS - proxy only ####
 

litellm/constants.py

@@ -158,6 +158,8 @@ DEFAULT_SEMANTIC_GUARD_EMBEDDING_MODEL: Final = str(
 )
 DEFAULT_SEMANTIC_GUARD_SIMILARITY_THRESHOLD = float(os.getenv("DEFAULT_SEMANTIC_GUARD_SIMILARITY_THRESHOLD", 0.75))
 
+DEFAULT_OPENAI_MODERATIONS_MODEL: Final = "omni-moderation-latest"
+
 # MCP OAuth2 Client Credentials Defaults
 MCP_OAUTH2_TOKEN_EXPIRY_BUFFER_SECONDS: Final = int(os.getenv("MCP_OAUTH2_TOKEN_EXPIRY_BUFFER_SECONDS", "60"))
 MCP_OAUTH2_TOKEN_CACHE_MAX_SIZE: Final = int(os.getenv("MCP_OAUTH2_TOKEN_CACHE_MAX_SIZE", "200"))

tests/test_litellm/proxy/guardrails/test_guardrail_coverage.py

@@ -18,7 +18,9 @@ from unittest.mock import AsyncMock, MagicMock, patch
 import pytest
 from httpx import Request, Response
 
+import litellm
 from litellm import DualCache
+from litellm.constants import DEFAULT_OPENAI_MODERATIONS_MODEL
 from litellm.proxy._types import UserAPIKeyAuth
 from litellm.types.utils import Choices, Message, ModelResponse
 
@@ -764,6 +766,40 @@ async def test_openai_moderation_inspects_multimodal_content(monkeypatch, user_a
     assert seen_inputs == ["alpha beta"]
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    ("configured_after_init", "expected_model"),
+    [("omni-moderation-2024-09-26", "omni-moderation-2024-09-26"), (None, DEFAULT_OPENAI_MODERATIONS_MODEL)],
+)
+async def test_openai_moderation_reads_model_name_at_call_time(
+    monkeypatch, user_api_key, configured_after_init, expected_model
+):
+    """``litellm_settings`` applies ``callbacks`` and ``openai_moderations_model_name`` in YAML
+    order, so the hook must resolve the model when it runs, not when it is constructed."""
+    from enterprise.enterprise_hooks.openai_moderation import (
+        _ENTERPRISE_OpenAI_Moderation,
+    )
+
+    monkeypatch.setattr(litellm, "openai_moderations_model_name", None)
+    guard = _ENTERPRISE_OpenAI_Moderation()
+    monkeypatch.setattr(litellm, "openai_moderations_model_name", configured_after_init)
+
+    class FakeModeration:
+        results = [type("R", (), {"flagged": False})()]
+
+    fake_router = MagicMock()
+    fake_router.amoderation = AsyncMock(return_value=FakeModeration())
+    monkeypatch.setattr("litellm.proxy.proxy_server.llm_router", fake_router, raising=False)
+
+    await guard.async_moderation_hook(
+        data={"messages": [{"role": "user", "content": "hello"}]},
+        user_api_key_dict=user_api_key,
+        call_type="acompletion",
+    )
+
+    fake_router.amoderation.assert_awaited_once_with(model=expected_model, input="hello")
+
+
 # ── Google Text Moderation ────────────────────────────────────────────────────