mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-12 08:42:18 +00:00
fix(router): clear type-check gate violations from previous-response reuse
This commit is contained in:
parent
daf924c0ef
commit
12db2c6b0d
2 changed files with 42 additions and 5 deletions
|
|
@ -3453,11 +3453,17 @@ class ComplexityRouter(CustomLogger):
|
|||
return default_model, self._tier_for_model(default_model)
|
||||
return None
|
||||
|
||||
@staticmethod
|
||||
def _deployment_model_id(deployment: Mapping[str, object]) -> str | None:
|
||||
model_info: Final = deployment.get("model_info")
|
||||
deployment_id: Final = cast( # cast-ok: the isinstance check in the condition narrows model_info to a Mapping with str keys
|
||||
"Mapping[str, object]", model_info
|
||||
).get("id") if isinstance(model_info, Mapping) else None
|
||||
return None if deployment_id is None else str(deployment_id)
|
||||
|
||||
def _group_has_deployment_id(self, group: str, model_id: str) -> bool:
|
||||
for deployment in self._group_deployments(group):
|
||||
model_info: Final = deployment.get("model_info")
|
||||
deployment_id: Final = model_info.get("id") if isinstance(model_info, Mapping) else None
|
||||
if deployment_id is not None and str(deployment_id) == model_id:
|
||||
if self._deployment_model_id(deployment) == model_id:
|
||||
return True
|
||||
return False
|
||||
|
||||
|
|
@ -4419,7 +4425,12 @@ class ComplexityRouter(CustomLogger):
|
|||
marker_pairs: Final = self._reminder_markers_for_request(request_kwargs)
|
||||
conversation_continuing: Final = (
|
||||
_conversation_is_continuing(resolved_messages)
|
||||
or _previous_response_id_from_request_kwargs(request_kwargs) is not None
|
||||
or _previous_response_id_from_request_kwargs(
|
||||
cast( # cast-ok: the hook's bare kwargs dict is the str-keyed request mapping
|
||||
Mapping[str, object], request_kwargs
|
||||
)
|
||||
)
|
||||
is not None
|
||||
)
|
||||
|
||||
use_session_affinity: Final = self._uses_tier_pin
|
||||
|
|
@ -4732,7 +4743,11 @@ class ComplexityRouter(CustomLogger):
|
|||
classifier_images: Final = self._classifier_image_parts(resolved_messages)
|
||||
|
||||
if user_message is None and not classifier_images:
|
||||
previous_response: Final = self._previous_response_model(request_kwargs)
|
||||
previous_response: Final = self._previous_response_model(
|
||||
cast( # cast-ok: the hook's bare kwargs dict is the str-keyed request mapping
|
||||
Mapping[str, object], request_kwargs
|
||||
)
|
||||
)
|
||||
if previous_response is not None:
|
||||
reused_model, reused_tier = previous_response
|
||||
reused_params: Final = self._litellm_params_for_model(reused_tier, reused_model)
|
||||
|
|
|
|||
|
|
@ -17404,6 +17404,28 @@ class TestPreviousResponseReuse:
|
|||
assert decision["tier"] == "COMPLEX", decision
|
||||
assert decision["conversation_continuing"] is True, decision
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_tool_continuation_reuses_default_model_when_only_it_served_the_previous_response(
|
||||
self, mock_router_instance, basic_config
|
||||
):
|
||||
router = self._router(
|
||||
mock_router_instance,
|
||||
basic_config,
|
||||
tiers={"SIMPLE": "gpt-4o-mini"},
|
||||
default_model="gpt-4o",
|
||||
)
|
||||
result = await router.async_pre_routing_hook(
|
||||
model="test-complexity-router",
|
||||
request_kwargs={"previous_response_id": self._encoded_response_id("deployment-4o")},
|
||||
messages=[{"role": "tool", "content": "search results", "tool_call_id": "call_1"}],
|
||||
)
|
||||
assert result is not None
|
||||
assert result.model == "gpt-4o", result.routing_decision
|
||||
decision = result.routing_decision
|
||||
assert decision is not None
|
||||
assert decision["cause"] == "previous_response_reuse", decision
|
||||
assert decision["conversation_continuing"] is True, decision
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_unencoded_previous_response_id_still_falls_back_to_default(
|
||||
self, mock_router_instance, basic_config
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue