Merge pull request #5254 from BerriAI/litellm_log_model_price_information

s3 - Log model price information

litellm/cost_calculator.py

@@ -410,6 +410,35 @@ def get_replicate_completion_pricing(completion_response=None, total_time=0.0):
     return a100_80gb_price_per_second_public * total_time / 1000
 
 
+def _select_model_name_for_cost_calc(
+    model: Optional[str],
+    completion_response: Union[BaseModel, dict],
+    base_model: Optional[str] = None,
+    custom_pricing: Optional[bool] = None,
+) -> Optional[str]:
+    """
+    1. If custom pricing is true, return received model name
+    2. If base_model is set (e.g. for azure models), return that
+    3. If completion response has model set return that
+    4. If model is passed in return that
+    """
+    if custom_pricing is True:
+        return model
+
+    if base_model is not None:
+        return base_model
+
+    return_model = model or completion_response.get("model", "")  # type: ignore
+    if hasattr(completion_response, "_hidden_params"):
+        if (
+            completion_response._hidden_params.get("model", None) is not None
+            and len(completion_response._hidden_params["model"]) > 0
+        ):
+            return_model = completion_response._hidden_params.get("model", model)
+
+    return return_model
+
+
 def completion_cost(
     completion_response=None,
     model: Optional[str] = None,
@@ -511,15 +540,10 @@ def completion_cost(
             verbose_logger.debug(
                 f"completion_response response ms: {getattr(completion_response, '_response_ms', None)} "
             )
-            model = model or completion_response.get(
-                "model", None
-            )  # check if user passed an override for model, if it's none check completion_response['model']
+            model = _select_model_name_for_cost_calc(
+                model=model, completion_response=completion_response
+            )
             if hasattr(completion_response, "_hidden_params"):
-                if (
-                    completion_response._hidden_params.get("model", None) is not None
-                    and len(completion_response._hidden_params["model"]) > 0
-                ):
-                    model = completion_response._hidden_params.get("model", model)
                 custom_llm_provider = completion_response._hidden_params.get(
                     "custom_llm_provider", custom_llm_provider or ""
                 )

litellm/litellm_core_utils/litellm_logging.py

@@ -24,6 +24,7 @@ from litellm import (
     verbose_logger,
 )
 from litellm.caching import DualCache, InMemoryCache, S3Cache
+from litellm.cost_calculator import _select_model_name_for_cost_calc
 from litellm.integrations.custom_logger import CustomLogger
 from litellm.litellm_core_utils.redact_messages import (
     redact_message_input_output_from_logging,
@@ -37,6 +38,7 @@ from litellm.types.utils import (
     ModelResponse,
     StandardLoggingHiddenParams,
     StandardLoggingMetadata,
+    StandardLoggingModelInformation,
     StandardLoggingPayload,
     TextCompletionResponse,
     TranscriptionResponse,
@@ -2293,6 +2295,38 @@ def get_standard_logging_object_payload(
 
             id = f"{id}_cache_hit{time.time()}"  # do not duplicate the request id
 
+        ## Get model cost information ##
+        base_model = _get_base_model_from_metadata(model_call_details=kwargs)
+        custom_pricing = use_custom_pricing_for_model(litellm_params=litellm_params)
+        model_cost_name = _select_model_name_for_cost_calc(
+            model=None,
+            completion_response=init_response_obj,
+            base_model=base_model,
+            custom_pricing=custom_pricing,
+        )
+        if model_cost_name is None:
+            model_cost_information = StandardLoggingModelInformation(
+                model_map_key="", model_map_value=None
+            )
+        else:
+            custom_llm_provider = kwargs.get("custom_llm_provider", None)
+
+            try:
+                _model_cost_information = litellm.get_model_info(
+                    model=model_cost_name, custom_llm_provider=custom_llm_provider
+                )
+                model_cost_information = StandardLoggingModelInformation(
+                    model_map_key=model_cost_name,
+                    model_map_value=_model_cost_information,
+                )
+            except Exception:
+                verbose_logger.warning(
+                    "Model is not mapped in model cost map. Defaulting to None model_cost_information for standard_logging_payload"
+                )
+                model_cost_information = StandardLoggingModelInformation(
+                    model_map_key=model_cost_name, model_map_value=None
+                )
+
         payload: StandardLoggingPayload = StandardLoggingPayload(
             id=str(id),
             call_type=call_type or "",
@@ -2319,6 +2353,7 @@ def get_standard_logging_object_payload(
             ),
             model_parameters=kwargs.get("optional_params", None),
             hidden_params=clean_hidden_params,
+            model_map_information=model_cost_information,
         )
 
         verbose_logger.debug(

litellm/proxy/_new_secret_config.yaml

@@ -3,4 +3,4 @@ model_list:
     litellm_params:
       model: "azure/azure-embedding-model"
       api_base: os.environ/AZURE_API_BASE
-      api_key: os.environ/AZURE_API_KEY
+      api_key: os.environ/AZURE_API_KEY

litellm/tests/test_completion_cost.py

@@ -251,7 +251,7 @@ def test_cost_azure_gpt_35():
         )
 
         cost = litellm.completion_cost(
-            completion_response=resp, model="azure/gpt-35-turbo"
+            completion_response=resp, model="azure/chatgpt-v-2"
         )
         print("\n Calculated Cost for azure/gpt-3.5-turbo", cost)
         input_cost = model_cost["azure/gpt-35-turbo"]["input_cost_per_token"]
@@ -262,9 +262,7 @@ def test_cost_azure_gpt_35():
         print("\n Excpected cost", expected_cost)
         assert cost == expected_cost
     except Exception as e:
-        pytest.fail(
-            f"Cost Calc failed for azure/gpt-3.5-turbo. Expected {expected_cost}, Calculated cost {cost}"
-        )
+        pytest.fail(f"Cost Calc failed for azure/gpt-3.5-turbo. {str(e)}")
 
 
 # test_cost_azure_gpt_35()

litellm/tests/test_custom_callback_input.py

@@ -1171,7 +1171,8 @@ def test_turn_off_message_logging():
 ##### VALID JSON ######
 
 
-def test_standard_logging_payload():
+@pytest.mark.parametrize("model", ["gpt-3.5-turbo", "azure/chatgpt-v-2"])
+def test_standard_logging_payload(model):
     """
     Ensure valid standard_logging_payload is passed for logging calls to s3
 
@@ -1187,9 +1188,9 @@ def test_standard_logging_payload():
         customHandler, "log_success_event", new=MagicMock()
     ) as mock_client:
         _ = litellm.completion(
-            model="gpt-3.5-turbo",
+            model=model,
             messages=[{"role": "user", "content": "Hey, how's it going?"}],
-            mock_response="Going well!",
+            # mock_response="Going well!",
         )
 
         time.sleep(2)
@@ -1204,7 +1205,11 @@ def test_standard_logging_payload():
             is not None
         )
 
-        print(mock_client.call_args.kwargs["kwargs"]["standard_logging_object"])
+        print(
+            "Standard Logging Object - {}".format(
+                mock_client.call_args.kwargs["kwargs"]["standard_logging_object"]
+            )
+        )
 
         keys_list = list(StandardLoggingPayload.__annotations__.keys())
 
@@ -1226,3 +1231,9 @@ def test_standard_logging_payload():
             ]
             > 0
         )
+        assert (
+            mock_client.call_args.kwargs["kwargs"]["standard_logging_object"][
+                "model_map_information"
+            ]["model_map_value"]
+            is not None
+        )

litellm/types/utils.py

@@ -1195,6 +1195,11 @@ class StandardLoggingHiddenParams(TypedDict):
     additional_headers: Optional[dict]
 
 
+class StandardLoggingModelInformation(TypedDict):
+    model_map_key: str
+    model_map_value: Optional[ModelInfo]
+
+
 class StandardLoggingPayload(TypedDict):
     id: str
     call_type: str
@@ -1205,6 +1210,7 @@ class StandardLoggingPayload(TypedDict):
     startTime: float
     endTime: float
     completionStartTime: float
+    model_map_information: StandardLoggingModelInformation
     model: str
     model_id: Optional[str]
     model_group: Optional[str]