scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-03 02:22:24 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 81

fix(router): assign savings metadata through typed keys

Browse source
This commit is contained in:
mikemikimike 2026-09-28 07:06:48 +08:00
parent 881b1a5e4a
commit 0f3b024e29
2 changed files with 15 additions and 24 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

18
litellm/router_strategy/adaptive_router/adaptive_router.py
View file

@ -222,26 +222,22 @@ class AdaptiveRouter:
request_type=request_type.value,
conversation_continuing=conversation_is_continuing(messages),
)
routing_decision.update(self._savings_baseline_fields())
self._add_savings_baseline_fields(routing_decision)
return PreRoutingHookResponse(
model=chosen_model,
messages=messages,
routing_decision=routing_decision,
)
def _savings_baseline_fields(self) -> StandardLoggingRoutingDecision:
def _add_savings_baseline_fields(self, decision: StandardLoggingRoutingDecision) -> None:
if self.litellm_router_instance is None:
return {} # mutable-ok: immutable empty result for absent router
return
baseline = resolve_baseline(self.litellm_router_instance, self.config.available_models)
if baseline is None:
return {} # mutable-ok: immutable empty result for unresolved baseline
fields: Final[StandardLoggingRoutingDecision] = {
"savings_baseline_model": baseline.model,
**(
{"savings_baseline_deployment_id": baseline.deployment_id} if baseline.deployment_id is not None else {}
),
}
return fields
return
decision["savings_baseline_model"] = baseline.model
if baseline.deployment_id is not None:
decision["savings_baseline_deployment_id"] = baseline.deployment_id
# ---- Pick model ------------------------------------------------------

21
litellm/router_strategy/quality_router/quality_router.py
View file

@ -323,17 +323,13 @@ class QualityRouter(CustomLogger):
if isinstance(metadata, dict):
metadata["quality_router_decision"] = decision
def _savings_fields(self) -> StandardLoggingRoutingDecision:
def _add_savings_fields(self, decision: StandardLoggingRoutingDecision) -> None:
baseline = resolve_baseline(self.litellm_router_instance, self.config.available_models)
if baseline is None:
return {} # mutable-ok: immutable empty result for unresolved baseline
fields: Final[StandardLoggingRoutingDecision] = {
"savings_baseline_model": baseline.model,
**(
{"savings_baseline_deployment_id": baseline.deployment_id} if baseline.deployment_id is not None else {}
),
}
return fields
return
decision["savings_baseline_model"] = baseline.model
if baseline.deployment_id is not None:
decision["savings_baseline_deployment_id"] = baseline.deployment_id
async def async_pre_routing_hook(
self,
@ -351,7 +347,6 @@ class QualityRouter(CustomLogger):
return None
conversation_continuing: Final = conversation_is_continuing(messages)
savings_fields: Final = self._savings_fields()
# Extract last user message and last system prompt — same rules as
# ComplexityRouter.async_pre_routing_hook.
@ -383,7 +378,7 @@ class QualityRouter(CustomLogger):
cause="default_fallback",
conversation_continuing=conversation_continuing,
)
default_routing_decision.update(savings_fields)
self._add_savings_fields(default_routing_decision)
return PreRoutingHookResponse(
model=self.config.default_model,
messages=messages,
@ -421,7 +416,7 @@ class QualityRouter(CustomLogger):
matched_keyword=matched_keyword,
conversation_continuing=conversation_continuing,
)
keyword_routing_decision.update(savings_fields)
self._add_savings_fields(keyword_routing_decision)
keyword_quality_tier: Final = self._model_quality.get(routed_model)
if keyword_quality_tier is not None:
keyword_routing_decision["tier"] = str(keyword_quality_tier)
@ -473,7 +468,7 @@ class QualityRouter(CustomLogger):
signals=list(signals), # mutable-ok: routing decision metadata uses a JSON list
conversation_continuing=conversation_continuing,
)
quality_routing_decision.update(savings_fields)
self._add_savings_fields(quality_routing_decision)
return PreRoutingHookResponse(
model=routed_model,
messages=messages,