mirror of
https://github.com/BerriAI/litellm.git
synced 2026-09-16 23:41:43 +00:00
fix(llm-as-a-judge): fix P1 code quality issues - validate weights/on_failure, guard pre_call, handle multimodal, move imports to module level, fix spurious finally logging
This commit is contained in:
Ishaan Jaffer
parent
af15eda10d
commit
da130d2e8e
1 changed files with 89 additions and 56 deletions
|
|
@ -2,8 +2,9 @@
|
|||
|
||||
import json
|
||||
from datetime import datetime
|
||||
from typing import TYPE_CHECKING, Any, Dict, List, Literal, Optional, Union, cast
|
||||
from typing import TYPE_CHECKING, Any, Dict, List, Literal, Optional, Union
|
||||
|
||||
import litellm
|
||||
from fastapi import HTTPException
|
||||
|
||||
from litellm._logging import verbose_logger
|
||||
|
|
@ -29,6 +30,36 @@ Return ONLY valid JSON in this exact format:
|
|||
"overall_score": <weighted average 0-100>
|
||||
}"""
|
||||
|
||||
_VALID_ON_FAILURE = frozenset({"block", "log"})
|
||||
|
||||
|
||||
def _extract_text_from_content(content: Any) -> str:
|
||||
"""Return plain text from a message content field (str or multimodal list)."""
|
||||
if isinstance(content, str):
|
||||
return content
|
||||
if isinstance(content, list):
|
||||
parts = []
|
||||
for part in content:
|
||||
if isinstance(part, dict) and part.get("type") == "text":
|
||||
parts.append(part.get("text", ""))
|
||||
return " ".join(parts)
|
||||
return ""
|
||||
|
||||
|
||||
def _get_litellm_param(
|
||||
litellm_params: "LitellmParams",
|
||||
guardrail: "Guardrail",
|
||||
key: str,
|
||||
default: Any = None,
|
||||
) -> Any:
|
||||
val = getattr(litellm_params, key, None)
|
||||
if val is not None:
|
||||
return val
|
||||
raw = guardrail.get("litellm_params")
|
||||
if isinstance(raw, dict) and key in raw:
|
||||
return raw[key]
|
||||
return default
|
||||
|
||||
|
||||
def _build_judge_prompt(
|
||||
criteria: List[Dict[str, Any]],
|
||||
|
|
@ -40,9 +71,9 @@ def _build_judge_prompt(
|
|||
for c in criteria
|
||||
)
|
||||
conversation = "\n".join(
|
||||
f'{m.get("role", "user").upper()}: {m.get("content", "")}'
|
||||
f'{m.get("role", "user").upper()}: {_extract_text_from_content(m.get("content", ""))}'
|
||||
for m in messages
|
||||
if isinstance(m.get("content"), str)
|
||||
if m.get("content") is not None
|
||||
)
|
||||
return (
|
||||
f"Criteria to evaluate:\n{criteria_block}\n\n"
|
||||
|
|
@ -52,7 +83,7 @@ def _build_judge_prompt(
|
|||
|
||||
|
||||
class LLMAsAJudgeGuardrail(CustomGuardrail):
|
||||
"""Post-call (and optionally pre-call) guardrail that judges response quality via an LLM."""
|
||||
"""Post-call guardrail that judges response quality via an LLM."""
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
|
|
@ -67,7 +98,6 @@ class LLMAsAJudgeGuardrail(CustomGuardrail):
|
|||
default_on: bool = False,
|
||||
**kwargs: Any,
|
||||
) -> None:
|
||||
# Normalize event_hook strings to enum values (matches block_code_execution pattern)
|
||||
_event_hook: Optional[Union[GuardrailEventHooks, List[GuardrailEventHooks]]] = (
|
||||
None
|
||||
)
|
||||
|
|
@ -86,10 +116,7 @@ class LLMAsAJudgeGuardrail(CustomGuardrail):
|
|||
|
||||
super().__init__(
|
||||
guardrail_name=guardrail_name,
|
||||
supported_event_hooks=[
|
||||
GuardrailEventHooks.pre_call,
|
||||
GuardrailEventHooks.post_call,
|
||||
],
|
||||
supported_event_hooks=[GuardrailEventHooks.post_call],
|
||||
event_hook=_event_hook or GuardrailEventHooks.post_call,
|
||||
default_on=default_on,
|
||||
**kwargs,
|
||||
|
|
@ -104,8 +131,6 @@ class LLMAsAJudgeGuardrail(CustomGuardrail):
|
|||
messages: List[Dict[str, Any]],
|
||||
response_text: str,
|
||||
) -> Dict[str, Any]:
|
||||
import litellm
|
||||
|
||||
judge_messages = [
|
||||
{"role": "system", "content": JUDGE_SYSTEM_PROMPT},
|
||||
{
|
||||
|
|
@ -130,18 +155,22 @@ class LLMAsAJudgeGuardrail(CustomGuardrail):
|
|||
input_type: Literal["request", "response"],
|
||||
logging_obj: Optional["LiteLLMLoggingObj"] = None,
|
||||
) -> GenericGuardrailAPIInputs:
|
||||
# Only evaluate post-call (response text). Fail open on pre-call.
|
||||
if input_type != "response":
|
||||
return inputs
|
||||
|
||||
texts = inputs.get("texts") or []
|
||||
response_text = " ".join(texts)
|
||||
if not response_text:
|
||||
return inputs
|
||||
|
||||
start_time = datetime.now()
|
||||
status: GuardrailStatus = "success"
|
||||
judge_result: Dict[str, Any] = {}
|
||||
|
||||
try:
|
||||
texts = inputs.get("texts") or []
|
||||
response_text = " ".join(texts)
|
||||
messages: List[Dict[str, Any]] = request_data.get("messages") or []
|
||||
|
||||
if not response_text:
|
||||
return inputs
|
||||
|
||||
try:
|
||||
judge_result = await self._run_judge(messages, response_text)
|
||||
except Exception as judge_err:
|
||||
|
|
@ -150,29 +179,34 @@ class LLMAsAJudgeGuardrail(CustomGuardrail):
|
|||
)
|
||||
return inputs
|
||||
|
||||
overall_score = float(judge_result.get("overall_score", 100))
|
||||
try:
|
||||
overall_score = max(
|
||||
0.0, min(100.0, float(judge_result.get("overall_score", 100)))
|
||||
)
|
||||
except (TypeError, ValueError):
|
||||
verbose_logger.warning(
|
||||
"llm_as_a_judge: invalid overall_score from judge, failing open"
|
||||
)
|
||||
return inputs
|
||||
|
||||
passed = overall_score >= self.overall_threshold
|
||||
|
||||
# Write judge result to eval_information so EvalViewer on the logs page can render it
|
||||
eval_info: "StandardLoggingEvalInformation" = {
|
||||
"eval_name": self.guardrail_name,
|
||||
"overall_score": overall_score,
|
||||
"passed": passed,
|
||||
"judge_model": self.judge_model,
|
||||
"threshold": self.overall_threshold,
|
||||
"verdicts": judge_result.get("verdicts", []),
|
||||
}
|
||||
_metadata = request_data.setdefault("metadata", {})
|
||||
_eval_info = cast(
|
||||
"StandardLoggingEvalInformation",
|
||||
{
|
||||
"eval_name": self.guardrail_name,
|
||||
"overall_score": overall_score,
|
||||
"passed": passed,
|
||||
"judge_model": self.judge_model,
|
||||
"threshold": self.overall_threshold,
|
||||
"verdicts": judge_result.get("verdicts", []),
|
||||
},
|
||||
)
|
||||
existing = _metadata.get("eval_information")
|
||||
if isinstance(existing, list):
|
||||
existing.append(_eval_info)
|
||||
existing.append(eval_info)
|
||||
elif existing is not None:
|
||||
_metadata["eval_information"] = [existing, _eval_info]
|
||||
_metadata["eval_information"] = [existing, eval_info]
|
||||
else:
|
||||
_metadata["eval_information"] = _eval_info
|
||||
_metadata["eval_information"] = eval_info
|
||||
|
||||
if not passed and self.on_failure == "block":
|
||||
status = "guardrail_intervened"
|
||||
|
|
@ -189,17 +223,11 @@ class LLMAsAJudgeGuardrail(CustomGuardrail):
|
|||
return inputs
|
||||
|
||||
except HTTPException:
|
||||
status = "guardrail_intervened"
|
||||
raise
|
||||
except Exception as e:
|
||||
verbose_logger.warning(f"llm_as_a_judge guardrail unexpected error: {e}")
|
||||
return inputs
|
||||
finally:
|
||||
event_type = (
|
||||
GuardrailEventHooks.post_call
|
||||
if input_type == "response"
|
||||
else GuardrailEventHooks.pre_call
|
||||
)
|
||||
self.add_standard_logging_guardrail_information_to_request_data(
|
||||
guardrail_provider="llm_as_a_judge",
|
||||
guardrail_json_response=judge_result,
|
||||
|
|
@ -207,7 +235,7 @@ class LLMAsAJudgeGuardrail(CustomGuardrail):
|
|||
guardrail_status=status,
|
||||
start_time=start_time.timestamp(),
|
||||
end_time=datetime.now().timestamp(),
|
||||
event_type=event_type,
|
||||
event_type=GuardrailEventHooks.post_call,
|
||||
)
|
||||
|
||||
|
||||
|
|
@ -215,34 +243,37 @@ def initialize_guardrail(
|
|||
litellm_params: "LitellmParams",
|
||||
guardrail: "Guardrail",
|
||||
) -> LLMAsAJudgeGuardrail:
|
||||
import litellm
|
||||
|
||||
def _get(key: str, default: Any = None) -> Any:
|
||||
val = getattr(litellm_params, key, None)
|
||||
if val is not None:
|
||||
return val
|
||||
raw = guardrail.get("litellm_params")
|
||||
if isinstance(raw, dict) and key in raw:
|
||||
return raw[key]
|
||||
return default
|
||||
|
||||
guardrail_name = guardrail.get("guardrail_name")
|
||||
if not guardrail_name:
|
||||
raise ValueError("llm_as_a_judge guardrail requires a guardrail_name")
|
||||
|
||||
judge_model = _get("judge_model")
|
||||
judge_model = _get_litellm_param(litellm_params, guardrail, "judge_model")
|
||||
if not judge_model:
|
||||
raise ValueError(
|
||||
"llm_as_a_judge guardrail requires judge_model in litellm_params"
|
||||
)
|
||||
|
||||
criteria = _get("criteria") or []
|
||||
criteria = _get_litellm_param(litellm_params, guardrail, "criteria") or []
|
||||
if not criteria:
|
||||
raise ValueError("llm_as_a_judge guardrail requires at least one criterion")
|
||||
|
||||
overall_threshold = float(_get("overall_threshold", 80.0))
|
||||
on_failure = _get("on_failure", "block")
|
||||
mode = _get("mode")
|
||||
weight_total = sum(int(c.get("weight", 0)) for c in criteria)
|
||||
if weight_total != 100:
|
||||
raise ValueError(
|
||||
f"llm_as_a_judge criterion weights must sum to 100 (got {weight_total})"
|
||||
)
|
||||
|
||||
on_failure = _get_litellm_param(litellm_params, guardrail, "on_failure", "block")
|
||||
if on_failure not in _VALID_ON_FAILURE:
|
||||
raise ValueError(
|
||||
f"llm_as_a_judge on_failure must be 'block' or 'log', got '{on_failure}'"
|
||||
)
|
||||
|
||||
overall_threshold = float(
|
||||
_get_litellm_param(litellm_params, guardrail, "overall_threshold", 80.0)
|
||||
)
|
||||
|
||||
mode = _get_litellm_param(litellm_params, guardrail, "mode")
|
||||
event_hook: Optional[GuardrailEventHooks] = None
|
||||
if isinstance(mode, str) and mode in {e.value for e in GuardrailEventHooks}:
|
||||
event_hook = GuardrailEventHooks(mode)
|
||||
|
|
@ -254,7 +285,9 @@ def initialize_guardrail(
|
|||
overall_threshold=overall_threshold,
|
||||
on_failure=on_failure,
|
||||
event_hook=event_hook,
|
||||
default_on=bool(_get("default_on", False)),
|
||||
default_on=bool(
|
||||
_get_litellm_param(litellm_params, guardrail, "default_on", False)
|
||||
),
|
||||
)
|
||||
litellm.logging_callback_manager.add_litellm_callback(instance)
|
||||
return instance
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue