Litellm dev 12 24 2024 p2 (BerriAI#7400)

* fix(utils.py): default custom_llm_provider=None for 'supports_response_schema'

Closes BerriAI#7397

* refactor(langfuse/): call langfuse logger inside customlogger compatible langfuse class, refactor langfuse logger to use verbose_logger.debug instead of print_verbose

* refactor(litellm_pre_call_utils.py): move config based team callbacks inside dynamic team callback logic

enables simpler unit testing for config-based team callbacks

* fix(proxy/_types.py): handle teamcallbackmetadata - none values

drop none values if present. if all none, use default dict to avoid downstream errors

* test(test_proxy_utils.py): add unit test preventing future issues - asserts team_id in config state not popped off across calls

Fixes BerriAI#6787

* fix(langfuse_prompt_management.py): add success + failure logging event support

* fix: fix linting error

* test: fix test

* test: fix test

* test: override o1 prompt caching - openai currently not working

* test: fix test

litellm/integrations/langfuse/langfuse.py

@@ -148,12 +148,7 @@ def add_metadata_from_header(litellm_params: dict, metadata: dict) -> dict:
 
         return metadata
 
-    # def log_error(kwargs, response_obj, start_time, end_time):
-    #     generation = trace.generation(
-    #         level ="ERROR" # can be any of DEBUG, DEFAULT, WARNING or ERROR
-    #         status_message='error' # can be any string (e.g. stringified stack trace or error body)
-    #     )
-    def log_event(  # noqa: PLR0915
+    def _old_log_event(  # noqa: PLR0915
         self,
         kwargs,
         response_obj,
@@ -167,7 +162,7 @@ def log_event(  # noqa: PLR0915
         # Method definition
 
         try:
-            print_verbose(
+            verbose_logger.debug(
                 f"Langfuse Logging - Enters logging function for model {kwargs}"
             )
 
@@ -260,7 +255,9 @@ def log_event(  # noqa: PLR0915
             ):
                 input = prompt
                 output = response_obj.get("response", "")
-            print_verbose(f"OUTPUT IN LANGFUSE: {output}; original: {response_obj}")
+            verbose_logger.debug(
+                f"OUTPUT IN LANGFUSE: {output}; original: {response_obj}"
+            )
             trace_id = None
             generation_id = None
             if self._is_langfuse_v2():
@@ -291,7 +288,7 @@ def log_event(  # noqa: PLR0915
                     input,
                     response_obj,
                 )
-            print_verbose(
+            verbose_logger.debug(
                 f"Langfuse Layer Logging - final response object: {response_obj}"
             )
             verbose_logger.info("Langfuse Layer Logging - logging success")
@@ -444,7 +441,7 @@ def _log_langfuse_v2(  # noqa: PLR0915
     ) -> tuple:
         import langfuse
 
-        print_verbose("Langfuse Layer Logging - logging to langfuse v2")
+        verbose_logger.debug("Langfuse Layer Logging - logging to langfuse v2")
 
         try:
             metadata = self._prepare_metadata(metadata)
@@ -577,7 +574,7 @@ def _log_langfuse_v2(  # noqa: PLR0915
                     trace_params["metadata"] = {"metadata_passed_to_litellm": metadata}
 
             cost = kwargs.get("response_cost", None)
-            print_verbose(f"trace: {cost}")
+            verbose_logger.debug(f"trace: {cost}")
 
             clean_metadata["litellm_response_cost"] = cost
             if standard_logging_object is not None:

litellm/integrations/langfuse/langfuse_prompt_management.py

@@ -14,7 +14,9 @@
 from litellm.integrations.custom_logger import CustomLogger
 from litellm.proxy._types import UserAPIKeyAuth
 from litellm.types.llms.openai import AllMessageValues
-from litellm.types.utils import StandardCallbackDynamicParams
+from litellm.types.utils import StandardCallbackDynamicParams, StandardLoggingPayload
+
+from .langfuse import LangFuseLogger
 
 if TYPE_CHECKING:
     from langfuse import Langfuse
@@ -92,7 +94,7 @@ def langfuse_client_init(
     return client
 
 
-class LangfusePromptManagement(CustomLogger):
+class LangfusePromptManagement(LangFuseLogger, CustomLogger):
     def __init__(
         self,
         langfuse_public_key=None,
@@ -248,3 +250,31 @@ def get_chat_completion_prompt(
         model = self._get_model_from_prompt(langfuse_prompt_client, model)
 
         return model, messages, non_default_params
+
+    async def async_log_success_event(self, kwargs, response_obj, start_time, end_time):
+        self._old_log_event(
+            kwargs=kwargs,
+            response_obj=response_obj,
+            start_time=start_time,
+            end_time=end_time,
+            user_id=kwargs.get("user", None),
+            print_verbose=None,
+        )
+
+    async def async_log_failure_event(self, kwargs, response_obj, start_time, end_time):
+        standard_logging_object = cast(
+            Optional[StandardLoggingPayload],
+            kwargs.get("standard_logging_object", None),
+        )
+        if standard_logging_object is None:
+            return
+        self._old_log_event(
+            start_time=start_time,
+            end_time=end_time,
+            response_obj=None,
+            user_id=kwargs.get("user", None),
+            print_verbose=None,
+            status_message=standard_logging_object["error_str"],
+            level="ERROR",
+            kwargs=kwargs,
+        )

litellm/litellm_core_utils/litellm_logging.py

@@ -1202,7 +1202,7 @@ def success_handler(  # noqa: PLR0915
                             in_memory_dynamic_logger_cache=in_memory_dynamic_logger_cache,
                         )
                         if langfuse_logger_to_use is not None:
-                            _response = langfuse_logger_to_use.log_event(
+                            _response = langfuse_logger_to_use._old_log_event(
                                 kwargs=kwargs,
                                 response_obj=result,
                                 start_time=start_time,
@@ -1925,7 +1925,7 @@ def failure_handler(  # noqa: PLR0915
                             standard_callback_dynamic_params=self.standard_callback_dynamic_params,
                             in_memory_dynamic_logger_cache=in_memory_dynamic_logger_cache,
                         )
-                        _response = langfuse_logger_to_use.log_event(
+                        _response = langfuse_logger_to_use._old_log_event(
                             start_time=start_time,
                             end_time=end_time,
                             response_obj=None,