(bug fix) - don't log messages, prompt, input in model_parameters in StandardLoggingPayload (#8923)

* fix _get_model_parameters

* test litellm logging

* test litellm logging

litellm/litellm_core_utils/litellm_logging.py

@@ -48,6 +48,7 @@ from litellm.types.router import SPECIAL_MODEL_INFO_PARAMS
 from litellm.types.utils import (
     CallTypes,
     EmbeddingResponse,
+    FieldsWithMessageContent,
     ImageResponse,
     LiteLLMLoggingBaseClass,
     ModelResponse,
@@ -3160,6 +3161,32 @@ class StandardLoggingPayloadSetup:
         else:
             return end_time_float - start_time_float
 
+    @staticmethod
+    def _remove_message_content_from_dict(original_dict: Optional[dict]) -> dict:
+        """
+        Filters out any params with message content `messages`, `input`, `prompt`
+
+        eg. We don't want to log the prompt in the model parameters
+        """
+        if original_dict is None:
+            return {}
+        sensitive_keys = FieldsWithMessageContent.get_all_fields()
+        cleaned_optional_params = {}
+        for key in original_dict:
+            if key not in sensitive_keys:
+                cleaned_optional_params[key] = original_dict[key]
+        return cleaned_optional_params
+
+    @staticmethod
+    def _get_model_parameters(kwargs: dict) -> dict:
+        """
+        Get the model parameters from the kwargs
+        """
+        optional_params = kwargs.get("optional_params", {}) or {}
+        return StandardLoggingPayloadSetup._remove_message_content_from_dict(
+            optional_params
+        )
+
 
 def get_standard_logging_object_payload(
     kwargs: Optional[dict],
@@ -3330,7 +3357,7 @@ def get_standard_logging_object_payload(
             requester_ip_address=clean_metadata.get("requester_ip_address", None),
             messages=kwargs.get("messages"),
             response=final_response_obj,
-            model_parameters=kwargs.get("optional_params", None),
+            model_parameters=StandardLoggingPayloadSetup._get_model_parameters(kwargs),
             hidden_params=clean_hidden_params,
             model_map_information=model_cost_information,
             error_str=error_str,

litellm/types/utils.py

@@ -1963,3 +1963,14 @@ class ProviderSpecificHeader(TypedDict):
 class SelectTokenizerResponse(TypedDict):
     type: Literal["openai_tokenizer", "huggingface_tokenizer"]
     tokenizer: Any
+
+
+class FieldsWithMessageContent(str, Enum):
+    MESSAGES = "messages"
+    INPUT = "input"
+    PROMPT = "prompt"
+    QUERY = "query"
+
+    @classmethod
+    def get_all_fields(cls) -> List[str]:
+        return [field.value for field in cls]

tests/litellm/litellm_core_utils/test_litellm_logging.py

@@ -0,0 +1,104 @@
+import json
+import os
+import sys
+
+import pytest
+from fastapi.testclient import TestClient
+
+sys.path.insert(
+    0, os.path.abspath("../../..")
+)  # Adds the parent directory to the system path
+
+from litellm.litellm_core_utils.litellm_logging import StandardLoggingPayloadSetup
+
+
+def test_remove_message_content_from_dict():
+    # Test with None input
+    assert StandardLoggingPayloadSetup._remove_message_content_from_dict(None) == {}
+
+    # Test with empty dict
+    assert StandardLoggingPayloadSetup._remove_message_content_from_dict({}) == {}
+
+    # Test with sensitive content
+    input_dict = {
+        "messages": "sensitive content",
+        "input": "secret prompt",
+        "prompt": "confidential text",
+        "safe_key": "safe value",
+        "temperature": 0.7,
+    }
+
+    expected_output = {"safe_key": "safe value", "temperature": 0.7}
+
+    result = StandardLoggingPayloadSetup._remove_message_content_from_dict(input_dict)
+    assert result == expected_output
+
+
+def test_get_model_parameters():
+    # Test with empty kwargs
+    assert StandardLoggingPayloadSetup._get_model_parameters({}) == {}
+
+    # Test with None optional_params
+    assert (
+        StandardLoggingPayloadSetup._get_model_parameters({"optional_params": None})
+        == {}
+    )
+
+    # Test with actual parameters
+    kwargs = {
+        "optional_params": {
+            "temperature": 0.8,
+            "messages": "sensitive data",
+            "max_tokens": 100,
+            "prompt": "secret prompt",
+        }
+    }
+
+    expected_output = {"temperature": 0.8, "max_tokens": 100}
+
+    result = StandardLoggingPayloadSetup._get_model_parameters(kwargs)
+    assert result == expected_output
+
+
+def test_get_model_parameters_complex():
+    # Test with more complex optional parameters
+    kwargs = {
+        "optional_params": {
+            "temperature": 0.8,
+            "messages": [{"role": "user", "content": "sensitive"}],
+            "max_tokens": 100,
+            "stop": ["\n", "stop"],
+            "presence_penalty": 0.5,
+            "frequency_penalty": 0.3,
+            "prompt": "secret prompt",
+            "n": 1,
+            "best_of": 2,
+            "logit_bias": {"50256": -100},
+        }
+    }
+
+    expected_output = {
+        "temperature": 0.8,
+        "max_tokens": 100,
+        "stop": ["\n", "stop"],
+        "presence_penalty": 0.5,
+        "frequency_penalty": 0.3,
+        "n": 1,
+        "best_of": 2,
+        "logit_bias": {"50256": -100},
+    }
+
+    result = StandardLoggingPayloadSetup._get_model_parameters(kwargs)
+    assert result == expected_output
+
+
+def test_get_model_parameters_empty_optional_params():
+    # Test with empty optional_params dictionary
+    kwargs = {"optional_params": {}}
+    assert StandardLoggingPayloadSetup._get_model_parameters(kwargs) == {}
+
+
+def test_get_model_parameters_missing_optional_params():
+    # Test with missing optional_params key
+    kwargs = {"model": "gpt-4", "api_key": "test"}
+    assert StandardLoggingPayloadSetup._get_model_parameters(kwargs) == {}