[Fix] o1-mini causes pydantic warnings on reasoning_tokens (#5754)

* add requester_metadata in standard logging payload

* log requester_metadata in metadata

* use StandardLoggingPayload for logging

* docs StandardLoggingPayload

* fix import

* include standard logging object in failure

* add test for requester metadata

* handle completion_tokens_details

* add test for completion_tokens_details

docs/my-website/docs/proxy/logging.md

@@ -61,7 +61,7 @@ litellm_settings:
 
 Removes any field with `user_api_key_*` from metadata.
 
-## What gets logged?
+## What gets logged? StandardLoggingPayload
 
 Found under `kwargs["standard_logging_object"]`. This is a standard payload, logged for every response.
 

litellm/integrations/gcs_bucket.py

@@ -16,6 +16,7 @@ from litellm.litellm_core_utils.logging_utils import (
 )
 from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler
 from litellm.proxy._types import CommonProxyErrors, SpendLogsMetadata, SpendLogsPayload
+from litellm.types.utils import StandardLoggingMetadata, StandardLoggingPayload
 
 
 class RequestKwargs(TypedDict):
@@ -30,6 +31,7 @@ class GCSBucketPayload(TypedDict):
     start_time: str
     end_time: str
     response_cost: Optional[float]
+    metadata: Optional[StandardLoggingMetadata]
     spend_log_metadata: str
     exception: Optional[str]
     log_event_type: Optional[str]
@@ -183,13 +185,22 @@ class GCSBucketLogger(GCSBucketBase):
             end_user_id=kwargs.get("end_user_id", None),
         )
 
+        # Ensure everything in the payload is converted to str
+        payload: Optional[StandardLoggingPayload] = kwargs.get(
+            "standard_logging_object", None
+        )
+
+        if payload is None:
+            raise ValueError("standard_logging_object not found in kwargs")
+
         gcs_payload: GCSBucketPayload = GCSBucketPayload(
             request_kwargs=request_kwargs,
             response_obj=response_dict,
             start_time=start_time,
             end_time=end_time,
+            metadata=payload["metadata"],
             spend_log_metadata=_spend_log_payload.get("metadata", ""),
-            response_cost=kwargs.get("response_cost", None),
+            response_cost=payload["response_cost"],
             exception=exception_str,
             log_event_type=None,
         )

litellm/litellm_core_utils/litellm_logging.py

@@ -1628,6 +1628,17 @@ class Logging:
         self.model_call_details.setdefault("original_response", None)
         self.model_call_details["response_cost"] = 0
 
+        ## STANDARDIZED LOGGING PAYLOAD
+        self.model_call_details["standard_logging_object"] = (
+            get_standard_logging_object_payload(
+                kwargs=self.model_call_details,
+                init_response_obj={},
+                start_time=start_time,
+                end_time=end_time,
+                logging_obj=self,
+            )
+        )
+
         if hasattr(exception, "headers") and isinstance(exception.headers, dict):
             self.model_call_details.setdefault("litellm_params", {})
             metadata = (
@@ -2419,6 +2430,7 @@ def get_standard_logging_object_payload(
             user_api_key_team_alias=None,
             spend_logs_metadata=None,
             requester_ip_address=None,
+            requester_metadata=None,
         )
         if isinstance(metadata, dict):
             # Filter the metadata dictionary to include only the specified keys

litellm/proxy/litellm_pre_call_utils.py

@@ -96,7 +96,7 @@ def convert_key_logging_metadata_to_callback(
     for var, value in data.callback_vars.items():
         if team_callback_settings_obj.callback_vars is None:
             team_callback_settings_obj.callback_vars = {}
-        team_callback_settings_obj.callback_vars[var] = (
+        team_callback_settings_obj.callback_vars[var] = str(
             litellm.utils.get_secret(value, default_value=value) or value
         )
 
@@ -204,6 +204,13 @@ async def add_litellm_data_to_request(
 
     if _metadata_variable_name not in data:
         data[_metadata_variable_name] = {}
+
+    # We want to log the "metadata" from the client side request. Avoid circular reference by not directly assigning metadata to itself
+    if "metadata" in data and data["metadata"] is not None:
+        data[_metadata_variable_name]["requester_metadata"] = copy.deepcopy(
+            data["metadata"]
+        )
+
     data[_metadata_variable_name]["user_api_key"] = user_api_key_dict.api_key
     data[_metadata_variable_name]["user_api_key_alias"] = getattr(
         user_api_key_dict, "key_alias", None

litellm/proxy/proxy_config.yaml

@@ -31,9 +31,8 @@ general_settings:
       "os.environ/SLACK_WEBHOOK_URL_2",
     ],
   }
-  key_management_system: "azure_key_vault"
 
 
 litellm_settings:
-  success_callback: ["prometheus"]
+  callbacks: ["gcs_bucket"]
 

litellm/tests/test_gcs_bucket.py

@@ -89,6 +89,7 @@ async def test_basic_gcs_logger():
             "user_api_key_team_alias": None,
             "user_api_key_metadata": {},
             "requester_ip_address": "127.0.0.1",
+            "requester_metadata": {"foo": "bar"},
             "spend_logs_metadata": {"hello": "world"},
             "headers": {
                 "content-type": "application/json",
@@ -159,6 +160,8 @@ async def test_basic_gcs_logger():
         == "116544810872468347480"
     )
 
+    assert gcs_payload["metadata"]["requester_metadata"] == {"foo": "bar"}
+
     # Delete Object from GCS
     print("deleting object from GCS")
     await gcs_logger.delete_gcs_object(object_name=object_name)

litellm/types/utils.py

@@ -5,7 +5,7 @@ from enum import Enum
 from typing import Any, Dict, List, Literal, Optional, Tuple, Union
 
 from openai._models import BaseModel as OpenAIObject
-from openai.types.audio.transcription_create_params import FileTypes
+from openai.types.audio.transcription_create_params import FileTypes  # type: ignore
 from openai.types.completion_usage import CompletionTokensDetails, CompletionUsage
 from pydantic import ConfigDict, Field, PrivateAttr
 from typing_extensions import Callable, Dict, Required, TypedDict, override
@@ -253,7 +253,7 @@ class HiddenParams(OpenAIObject):
         # Allow dictionary-style assignment of attributes
         setattr(self, key, value)
 
-    def json(self, **kwargs):
+    def json(self, **kwargs):  # type: ignore
         try:
             return self.model_dump()  # noqa
         except:
@@ -359,7 +359,7 @@ class Message(OpenAIObject):
         # Allow dictionary-style assignment of attributes
         setattr(self, key, value)
 
-    def json(self, **kwargs):
+    def json(self, **kwargs):  # type: ignore
         try:
             return self.model_dump()  # noqa
         except:
@@ -490,6 +490,19 @@ class Usage(CompletionUsage):
             completion_tokens_details = CompletionTokensDetails(
                 reasoning_tokens=reasoning_tokens
             )
+
+        # Ensure completion_tokens_details is properly handled
+        if "completion_tokens_details" in params:
+            if isinstance(params["completion_tokens_details"], dict):
+                completion_tokens_details = CompletionTokensDetails(
+                    **params["completion_tokens_details"]
+                )
+            elif isinstance(
+                params["completion_tokens_details"], CompletionTokensDetails
+            ):
+                completion_tokens_details = params["completion_tokens_details"]
+            del params["completion_tokens_details"]
+
         super().__init__(
             prompt_tokens=prompt_tokens or 0,
             completion_tokens=completion_tokens or 0,
@@ -641,6 +654,7 @@ class ModelResponse(OpenAIObject):
             if choices is not None and isinstance(choices, list):
                 new_choices = []
                 for choice in choices:
+                    _new_choice = None
                     if isinstance(choice, StreamingChoices):
                         _new_choice = choice
                     elif isinstance(choice, dict):
@@ -715,7 +729,7 @@ class ModelResponse(OpenAIObject):
         # Allow dictionary-style access to attributes
         return getattr(self, key)
 
-    def json(self, **kwargs):
+    def json(self, **kwargs):  # type: ignore
         try:
             return self.model_dump()  # noqa
         except:
@@ -804,7 +818,7 @@ class EmbeddingResponse(OpenAIObject):
         # Allow dictionary-style assignment of attributes
         setattr(self, key, value)
 
-    def json(self, **kwargs):
+    def json(self, **kwargs):  # type: ignore
         try:
             return self.model_dump()  # noqa
         except:
@@ -855,7 +869,7 @@ class TextChoices(OpenAIObject):
         # Allow dictionary-style assignment of attributes
         setattr(self, key, value)
 
-    def json(self, **kwargs):
+    def json(self, **kwargs):  # type: ignore
         try:
             return self.model_dump()  # noqa
         except:
@@ -911,6 +925,7 @@ class TextCompletionResponse(OpenAIObject):
             if choices is not None and isinstance(choices, list):
                 new_choices = []
                 for choice in choices:
+                    _new_choice = None
                     if isinstance(choice, TextChoices):
                         _new_choice = choice
                     elif isinstance(choice, dict):
@@ -937,12 +952,12 @@ class TextCompletionResponse(OpenAIObject):
             usage = Usage()
 
         super(TextCompletionResponse, self).__init__(
-            id=id,
-            object=object,
-            created=created,
-            model=model,
-            choices=choices,
-            usage=usage,
+            id=id,  # type: ignore
+            object=object,  # type: ignore
+            created=created,  # type: ignore
+            model=model,  # type: ignore
+            choices=choices,  # type: ignore
+            usage=usage,  # type: ignore
             **params,
         )
 
@@ -986,7 +1001,7 @@ class ImageObject(OpenAIObject):
     revised_prompt: Optional[str] = None
 
     def __init__(self, b64_json=None, url=None, revised_prompt=None):
-        super().__init__(b64_json=b64_json, url=url, revised_prompt=revised_prompt)
+        super().__init__(b64_json=b64_json, url=url, revised_prompt=revised_prompt)  # type: ignore
 
     def __contains__(self, key):
         # Define custom behavior for the 'in' operator
@@ -1004,7 +1019,7 @@ class ImageObject(OpenAIObject):
         # Allow dictionary-style assignment of attributes
         setattr(self, key, value)
 
-    def json(self, **kwargs):
+    def json(self, **kwargs):  # type: ignore
         try:
             return self.model_dump()  # noqa
         except:
@@ -1057,7 +1072,7 @@ class ImageResponse(OpenAIImageResponse):
         # Allow dictionary-style assignment of attributes
         setattr(self, key, value)
 
-    def json(self, **kwargs):
+    def json(self, **kwargs):  # type: ignore
         try:
             return self.model_dump()  # noqa
         except:
@@ -1072,7 +1087,7 @@ class TranscriptionResponse(OpenAIObject):
     _response_headers: Optional[dict] = None
 
     def __init__(self, text=None):
-        super().__init__(text=text)
+        super().__init__(text=text)  # type: ignore
 
     def __contains__(self, key):
         # Define custom behavior for the 'in' operator
@@ -1090,7 +1105,7 @@ class TranscriptionResponse(OpenAIObject):
         # Allow dictionary-style assignment of attributes
         setattr(self, key, value)
 
-    def json(self, **kwargs):
+    def json(self, **kwargs):  # type: ignore
         try:
             return self.model_dump()  # noqa
         except:
@@ -1247,6 +1262,7 @@ class StandardLoggingMetadata(TypedDict):
         dict
     ]  # special param to log k,v pairs to spendlogs for a call
     requester_ip_address: Optional[str]
+    requester_metadata: Optional[dict]
 
 
 class StandardLoggingHiddenParams(TypedDict):

tests/llm_translation/test_openai_o1.py

@@ -99,3 +99,24 @@ async def test_o1_max_completion_tokens(respx_mock: MockRouter, model: str):
 
     print(f"response: {response}")
     assert isinstance(response, ModelResponse)
+
+
+def test_litellm_responses():
+    """
+    ensures that type of completion_tokens_details is correctly handled / returned
+    """
+    from litellm import ModelResponse
+    from litellm.types.utils import CompletionTokensDetails
+
+    response = ModelResponse(
+        usage={
+            "completion_tokens": 436,
+            "prompt_tokens": 14,
+            "total_tokens": 450,
+            "completion_tokens_details": {"reasoning_tokens": 0},
+        }
+    )
+
+    print("response: ", response)
+
+    assert isinstance(response.usage.completion_tokens_details, CompletionTokensDetails)