scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-10 22:41:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 83

test_openai_moderation_responses_api_input_field

Browse source
This commit is contained in:
Ishaan Jaffer 2026-02-05 13:37:55 -08:00
parent ff89e8d1af
commit 0e285559fa
1 changed files with 30 additions and 40 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

70
tests/local_testing/test_openai_moderations_hook.py
View file

@ -90,13 +90,12 @@ async def test_openai_moderation_error_raising(monkeypatch):
@pytest.mark.asyncio
async def test_openai_moderation_responses_api_input_field():
"""
Tests that OpenAI Moderation works with Responses API input field.
Tests that OpenAI Moderation works with Responses API input field via apply_guardrail.
This test verifies the fix for the issue where moderation was skipped
for Responses API because it only checked for 'messages' field but
Responses API uses 'input' field instead.
This test verifies that the unified guardrail interface (apply_guardrail) correctly
handles different input types: plain text strings, structured messages, and lists.
"""
from unittest.mock import AsyncMock, MagicMock, patch
from unittest.mock import patch
from litellm.types.llms.openai import (
OpenAIModerationResponse,
OpenAIModerationResult,
@ -104,6 +103,7 @@ async def test_openai_moderation_responses_api_input_field():
from litellm.proxy.guardrails.guardrail_hooks.openai.moderations import (
OpenAIModerationGuardrail,
)
from litellm.types.utils import GenericGuardrailAPIInputs
# Initialize the open-source OpenAI Moderation guardrail
openai_mod = OpenAIModerationGuardrail(
@ -112,10 +112,6 @@ async def test_openai_moderation_responses_api_input_field():
model="omni-moderation-latest",
)
_api_key = "sk-12345"
_api_key = hash_token("sk-12345")
user_api_key_dict = UserAPIKeyAuth(api_key=_api_key)
# Mock the async_make_request to return a flagged response
mock_moderation_response = OpenAIModerationResponse(
id="modr-123",
@ -133,53 +129,47 @@ async def test_openai_moderation_responses_api_input_field():
with patch.object(
openai_mod, "async_make_request", return_value=mock_moderation_response
):
# Test 1: Responses API with input as string
# Test 1: Responses API / Embeddings with texts (string input)
try:
await openai_mod.async_moderation_hook(
data={
"model": "gpt-4o",
"input": "I want to hurt people",
},
user_api_key_dict=user_api_key_dict,
call_type="responses",
inputs = GenericGuardrailAPIInputs(texts=["I want to hurt people"])
await openai_mod.apply_guardrail(
inputs=inputs,
request_data={"model": "gpt-4o", "input": "I want to hurt people"},
input_type="request",
)
pytest.fail("Should have raised HTTPException for flagged content")
except Exception as e:
print("Got exception for string input: ", e)
print("Got exception for texts input: ", e)
assert "Violated OpenAI moderation policy" in str(e)
# Test 2: Responses API with input as list of messages
# Test 2: Responses API with structured_messages (list of message objects)
try:
await openai_mod.async_moderation_hook(
data={
"model": "gpt-4o",
"input": [
{"role": "user", "content": "I want to hurt people"}
],
},
user_api_key_dict=user_api_key_dict,
call_type="responses",
inputs = GenericGuardrailAPIInputs(
structured_messages=[{"role": "user", "content": "I want to hurt people"}]
)
await openai_mod.apply_guardrail(
inputs=inputs,
request_data={"model": "gpt-4o", "input": [{"role": "user", "content": "I want to hurt people"}]},
input_type="request",
)
pytest.fail("Should have raised HTTPException for flagged content")
except Exception as e:
print("Got exception for list input: ", e)
print("Got exception for structured_messages input: ", e)
assert "Violated OpenAI moderation policy" in str(e)
# Test 3: Verify it still works with messages field (Chat Completions)
# Test 3: Chat Completions with structured_messages
try:
await openai_mod.async_moderation_hook(
data={
"model": "gpt-4o",
"messages": [
{"role": "user", "content": "I want to hurt people"}
],
},
user_api_key_dict=user_api_key_dict,
call_type="completion",
inputs = GenericGuardrailAPIInputs(
structured_messages=[{"role": "user", "content": "I want to hurt people"}]
)
await openai_mod.apply_guardrail(
inputs=inputs,
request_data={"model": "gpt-4o", "messages": [{"role": "user", "content": "I want to hurt people"}]},
input_type="request",
)
pytest.fail("Should have raised HTTPException for flagged content")
except Exception as e:
print("Got exception for messages field: ", e)
print("Got exception for chat completions input: ", e)
assert "Violated OpenAI moderation policy" in str(e)
print("✓ All Responses API moderation tests passed!")