scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-13 23:11:40 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 101

fix: merge HiddenParams into logging metadata; lazy-import response_metadata to avoid cycles

Browse source 
Made-with: Cursor
This commit is contained in:
shivam 2026-03-27 16:09:49 -07:00
parent ec736e5975
commit 76cb94157e
4 changed files with 86 additions and 20 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

42
litellm/litellm_core_utils/litellm_logging.py
View file

@ -60,9 +60,6 @@ from litellm.integrations.deepeval.deepeval import DeepEvalLogger
from litellm.integrations.mlflow import MlflowLogger
from litellm.integrations.sqs import SQSLogger
from litellm.litellm_core_utils.core_helpers import reconstruct_model_name
from litellm.litellm_core_utils.llm_response_utils.response_metadata import (
get_response_hidden_params,
)
from litellm.litellm_core_utils.get_litellm_params import get_litellm_params
from litellm.litellm_core_utils.llm_cost_calc.tool_call_cost_tracking import (
StandardBuiltInToolCostTracking,
@ -1705,7 +1702,17 @@ class Logging(LiteLLMLoggingBaseClass):
"""
if logging_result is None:
return
incoming = get_response_hidden_params(logging_result)
from litellm.litellm_core_utils.llm_response_utils.response_metadata import (
get_response_hidden_params,
hidden_params_to_plain_dict,
)
incoming_raw = get_response_hidden_params(logging_result)
incoming = hidden_params_to_plain_dict(incoming_raw)
if not incoming:
return
# Do not let explicit None values from model_dump wipe proxy-injected timing keys.
incoming = {k: v for k, v in incoming.items() if v is not None}
if not incoming:
return
if self.model_call_details.get("litellm_params") is None:
@ -1719,9 +1726,6 @@ class Logging(LiteLLMLoggingBaseClass):
existing = md.get("hidden_params")
if not isinstance(existing, dict):
existing = {}
if not isinstance(incoming, dict):
md["hidden_params"] = existing or incoming
return
md["hidden_params"] = {**existing, **incoming}
def _process_hidden_params_and_response_cost(
@ -1730,18 +1734,20 @@ class Logging(LiteLLMLoggingBaseClass):
start_time,
end_time,
):
hidden_params = get_response_hidden_params(logging_result)
if hidden_params:
if self.model_call_details.get("litellm_params") is not None:
self.model_call_details["litellm_params"].setdefault("metadata", {})
if self.model_call_details["litellm_params"]["metadata"] is None:
self.model_call_details["litellm_params"]["metadata"] = {}
self.model_call_details["litellm_params"]["metadata"]["hidden_params"] = get_response_hidden_params(logging_result) # type: ignore
from litellm.litellm_core_utils.llm_response_utils.response_metadata import (
get_response_hidden_params,
hidden_params_to_plain_dict,
)
hp_raw = get_response_hidden_params(logging_result)
if hp_raw:
self._merge_hidden_params_from_response_into_metadata(logging_result)
hp_dict = hidden_params_to_plain_dict(hp_raw) if hp_raw else {}
if self.model_call_details.get("cache_hit") is True:
self.model_call_details["response_cost"] = 0.0
elif "response_cost" in hidden_params:
self.model_call_details["response_cost"] = hidden_params["response_cost"]
elif "response_cost" in hp_dict:
self.model_call_details["response_cost"] = hp_dict["response_cost"]
elif self.model_call_details.get("response_cost") is not None:
# Preserve response_cost if already calculated (e.g., by pass-through
# handlers like Gemini/Vertex which call completion_cost directly)
@ -5316,6 +5322,10 @@ def _extract_response_obj_and_hidden_params(
hidden_params = getattr(init_response_obj, "_hidden_params", None)
elif isinstance(init_response_obj, dict):
response_obj = init_response_obj
from litellm.litellm_core_utils.llm_response_utils.response_metadata import (
get_response_hidden_params,
)
hp = get_response_hidden_params(init_response_obj)
if hp:
hidden_params = (

14
litellm/litellm_core_utils/llm_response_utils/response_metadata.py
View file

@ -33,6 +33,20 @@ def get_response_hidden_params(response: Any) -> Union[HiddenParams, dict]:
return {}
def hidden_params_to_plain_dict(hp: Any) -> dict:
"""
Normalize ``get_response_hidden_params`` output (``dict`` or ``HiddenParams``) to a
plain ``dict`` for merging into ``metadata['hidden_params']``.
"""
if not hp:
return {}
if isinstance(hp, dict):
return dict(hp)
if hasattr(hp, "model_dump"):
return hp.model_dump(exclude_none=True)
return {}
def strip_litellm_internal_keys_from_dict_response(response: Any) -> None:
"""Remove internal keys from dict API responses before JSON serialization."""
if isinstance(response, dict):

13
litellm/proxy/common_request_processing.py
View file

@ -36,10 +36,6 @@ from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLogging
from litellm.litellm_core_utils.llm_response_utils.get_headers import (
get_response_headers,
)
from litellm.litellm_core_utils.llm_response_utils.response_metadata import (
merge_hidden_params_with_logging_timings,
strip_litellm_internal_keys_from_dict_response,
)
from litellm.litellm_core_utils.safe_json_dumps import safe_dumps
from litellm.proxy._types import ProxyException, UserAPIKeyAuth
from litellm.proxy.auth.auth_utils import check_response_size_is_safe
@ -1042,6 +1038,10 @@ class ProxyBaseLLMRequestProcessing:
_exception_raised = False
try:
from litellm.litellm_core_utils.llm_response_utils.response_metadata import (
merge_hidden_params_with_logging_timings,
)
# Async generators (e.g. Anthropic /v1/messages SSE) have no _hidden_params;
# merge timing from logging_obj.model_call_details (llm_api_duration_ms, etc.).
hidden_params = merge_hidden_params_with_logging_timings(
@ -1287,6 +1287,11 @@ class ProxyBaseLLMRequestProcessing:
log_context=f"litellm_call_id={logging_obj.litellm_call_id}",
)
from litellm.litellm_core_utils.llm_response_utils.response_metadata import (
merge_hidden_params_with_logging_timings,
strip_litellm_internal_keys_from_dict_response,
)
# Re-merge after post_call_success_hook so dict / CSW hidden_params + logging timings align.
hidden_params = merge_hidden_params_with_logging_timings(
response,

37
tests/test_litellm/litellm_core_utils/test_litellm_logging.py
View file

@ -2333,6 +2333,43 @@ def test_merge_hidden_params_from_response_preserves_proxy_injected_timing():
assert hp["model_id"] == "mid-2"
def test_merge_hidden_params_from_response_preserves_proxy_injected_timing_with_hidden_params_model():
"""Assembled streaming response may use HiddenParams object; merge must not drop fields."""
from litellm.types.llms.base import HiddenParams
from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
logging_obj = LiteLLMLoggingObj(
model="claude-3-5-sonnet",
messages=[{"role": "user", "content": "hi"}],
stream=True,
call_type="anthropic_messages",
start_time=time.time(),
litellm_call_id="merge-hp-hp-model",
function_id="merge-hp-hp-model-fn",
)
logging_obj.model_call_details = {
"litellm_params": {
"metadata": {
"hidden_params": {
"litellm_overhead_time_ms": 273.329,
"_response_ms": 1590.0,
}
}
},
}
class _Assembled:
_hidden_params = HiddenParams(response_cost=0.001, model_id="mid-2")
logging_obj._merge_hidden_params_from_response_into_metadata(_Assembled())
hp = logging_obj.model_call_details["litellm_params"]["metadata"]["hidden_params"]
assert hp["litellm_overhead_time_ms"] == 273.329
assert hp["_response_ms"] == 1590.0
assert hp["response_cost"] == 0.001
assert hp["model_id"] == "mid-2"
def test_get_standard_logging_payload_includes_metadata_hidden_params_overhead():
"""Spend logs read overhead from standard_logging_object.hidden_params; merge metadata.hidden_params."""
from datetime import datetime