Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commit1bfd82b

Browse files
authored
Correctly passing API key to moderation (#36)
* extract call_moderation helper
1 parent45e958c commit1bfd82b

File tree

3 files changed

+128
-30
lines changed

3 files changed

+128
-30
lines changed

‎src/guardrails/checks/text/moderation.py‎

Lines changed: 39 additions & 30 deletions
Original file line numberDiff line numberDiff line change
@@ -32,7 +32,7 @@
3232
fromfunctoolsimportcache
3333
fromtypingimportAny
3434

35-
fromopenaiimportAsyncOpenAI
35+
fromopenaiimportAsyncOpenAI,NotFoundError
3636
frompydanticimportBaseModel,ConfigDict,Field
3737

3838
fromguardrails.registryimportdefault_spec_registry
@@ -132,6 +132,22 @@ def _get_moderation_client() -> AsyncOpenAI:
132132
returnAsyncOpenAI(**prepare_openai_kwargs({}))
133133

134134

135+
asyncdef_call_moderation_api(client:AsyncOpenAI,data:str)->Any:
136+
"""Call the OpenAI moderation API.
137+
138+
Args:
139+
client: The OpenAI client to use.
140+
data: The text to analyze.
141+
142+
Returns:
143+
The moderation API response.
144+
"""
145+
returnawaitclient.moderations.create(
146+
model="omni-moderation-latest",
147+
input=data,
148+
)
149+
150+
135151
asyncdefmoderation(
136152
ctx:Any,
137153
data:str,
@@ -151,36 +167,29 @@ async def moderation(
151167
Returns:
152168
GuardrailResult: Indicates if tripwire was triggered, and details of flagged categories.
153169
"""
154-
155-
# Prefer reusing an existing OpenAI client from context ONLY if it targets the
156-
# official OpenAI API. If it's any other provider (e.g., Ollama via base_url),
157-
# fall back to the default OpenAI moderation client.
158-
def_maybe_reuse_openai_client_from_ctx(context:Any)->AsyncOpenAI|None:
170+
client=None
171+
ifctxisnotNone:
172+
candidate=getattr(ctx,"guardrail_llm",None)
173+
ifisinstance(candidate,AsyncOpenAI):
174+
client=candidate
175+
176+
# Try the context client first, fall back if moderation endpoint doesn't exist
177+
ifclientisnotNone:
159178
try:
160-
candidate=getattr(context,"guardrail_llm",None)
161-
ifnotisinstance(candidate,AsyncOpenAI):
162-
returnNone
163-
164-
# Attempt to discover the effective base URL in a best-effort way
165-
base_url=getattr(candidate,"base_url",None)
166-
ifbase_urlisNone:
167-
inner=getattr(candidate,"_client",None)
168-
base_url=getattr(inner,"base_url",None)orgetattr(inner,"_base_url",None)
169-
170-
# Reuse only when clearly the official OpenAI endpoint
171-
ifbase_urlisNone:
172-
returncandidate
173-
ifisinstance(base_url,str)and"api.openai.com"inbase_url:
174-
returncandidate
175-
returnNone
176-
exceptException:
177-
returnNone
178-
179-
client=_maybe_reuse_openai_client_from_ctx(ctx)or_get_moderation_client()
180-
resp=awaitclient.moderations.create(
181-
model="omni-moderation-latest",
182-
input=data,
183-
)
179+
resp=await_call_moderation_api(client,data)
180+
exceptNotFoundErrorase:
181+
# Moderation endpoint doesn't exist on this provider (e.g., third-party)
182+
# Fall back to the OpenAI client
183+
logger.debug(
184+
"Moderation endpoint not available on context client, falling back to OpenAI: %s",
185+
e,
186+
)
187+
client=_get_moderation_client()
188+
resp=await_call_moderation_api(client,data)
189+
else:
190+
# No context client, use fallback
191+
client=_get_moderation_client()
192+
resp=await_call_moderation_api(client,data)
184193
results=resp.resultsor []
185194
ifnotresults:
186195
returnGuardrailResult(

‎tests/conftest.py‎

Lines changed: 11 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -67,7 +67,18 @@ class APITimeoutError(Exception):
6767
"""Stub API timeout error."""
6868

6969

70+
classNotFoundError(Exception):
71+
"""Stub 404 not found error."""
72+
73+
def__init__(self,message:str,*,response:Any=None,body:Any=None)->None:
74+
"""Initialize NotFoundError with OpenAI-compatible signature."""
75+
super().__init__(message)
76+
self.response=response
77+
self.body=body
78+
79+
7080
_STUB_OPENAI_MODULE.APITimeoutError=APITimeoutError
81+
_STUB_OPENAI_MODULE.NotFoundError=NotFoundError
7182

7283
_OPENAI_TYPES_MODULE=types.ModuleType("openai.types")
7384
_OPENAI_TYPES_MODULE.Completion=_DummyResponse

‎tests/unit/checks/test_moderation.py‎

Lines changed: 78 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -56,3 +56,81 @@ async def create_empty(**_: Any) -> Any:
5656

5757
assertresult.tripwire_triggeredisFalse# noqa: S101
5858
assertresult.info["error"]=="No moderation results returned"# noqa: S101
59+
60+
61+
@pytest.mark.asyncio
62+
asyncdeftest_moderation_uses_context_client()->None:
63+
"""Moderation should use the client from context when available."""
64+
fromopenaiimportAsyncOpenAI
65+
66+
# Track whether context client was used
67+
context_client_used=False
68+
69+
asyncdeftrack_create(**_:Any)->Any:
70+
nonlocalcontext_client_used
71+
context_client_used=True
72+
73+
class_Result:
74+
defmodel_dump(self)->dict[str,Any]:
75+
return {"categories": {"hate":False,"violence":False}}
76+
77+
returnSimpleNamespace(results=[_Result()])
78+
79+
# Create a context with a guardrail_llm client
80+
context_client=AsyncOpenAI(api_key="test-context-key",base_url="https://api.openai.com/v1")
81+
context_client.moderations=SimpleNamespace(create=track_create)# type: ignore[assignment]
82+
83+
ctx=SimpleNamespace(guardrail_llm=context_client)
84+
85+
cfg=ModerationCfg(categories=[Category.HATE])
86+
result=awaitmoderation(ctx,"test text",cfg)
87+
88+
# Verify the context client was used
89+
assertcontext_client_usedisTrue# noqa: S101
90+
assertresult.tripwire_triggeredisFalse# noqa: S101
91+
92+
93+
@pytest.mark.asyncio
94+
asyncdeftest_moderation_falls_back_for_third_party_provider(monkeypatch:pytest.MonkeyPatch)->None:
95+
"""Moderation should fall back to environment client for third-party providers."""
96+
fromopenaiimportAsyncOpenAI,NotFoundError
97+
98+
# Create fallback client that tracks usage
99+
fallback_used=False
100+
101+
asyncdeftrack_fallback_create(**_:Any)->Any:
102+
nonlocalfallback_used
103+
fallback_used=True
104+
105+
class_Result:
106+
defmodel_dump(self)->dict[str,Any]:
107+
return {"categories": {"hate":False}}
108+
109+
returnSimpleNamespace(results=[_Result()])
110+
111+
fallback_client=SimpleNamespace(moderations=SimpleNamespace(create=track_fallback_create))
112+
monkeypatch.setattr("guardrails.checks.text.moderation._get_moderation_client",lambda:fallback_client)
113+
114+
# Create a mock httpx.Response for NotFoundError
115+
mock_response=SimpleNamespace(
116+
status_code=404,
117+
headers={},
118+
text="404 page not found",
119+
json=lambda: {"error": {"message":"Not found","type":"invalid_request_error"}},
120+
)
121+
122+
# Create a context client that simulates a third-party provider
123+
# When moderation is called, it should raise NotFoundError
124+
asyncdefraise_not_found(**_:Any)->Any:
125+
raiseNotFoundError("404 page not found",response=mock_response,body=None)# type: ignore[arg-type]
126+
127+
third_party_client=AsyncOpenAI(api_key="third-party-key",base_url="https://localhost:8080/v1")
128+
third_party_client.moderations=SimpleNamespace(create=raise_not_found)# type: ignore[assignment]
129+
ctx=SimpleNamespace(guardrail_llm=third_party_client)
130+
131+
cfg=ModerationCfg(categories=[Category.HATE])
132+
result=awaitmoderation(ctx,"test text",cfg)
133+
134+
# Verify the fallback client was used (not the third-party one)
135+
assertfallback_usedisTrue# noqa: S101
136+
assertresult.tripwire_triggeredisFalse# noqa: S101

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp