phoenix/litellm
1
0
Fork 0
forked from phoenix/litellm-mirror
Code Pull requests Activity

fix(langfuse.py): serialize message for logging

Browse source
This commit is contained in:
Krrish Dholakia 2023-12-12 21:40:55 -08:00
parent 7856d65d44
commit f9dfeb502a
2 changed files with 13 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

2
litellm/integrations/langfuse.py
View file

@ -58,7 +58,7 @@ class LangFuseLogger:
model=kwargs['model'], model=kwargs['model'],
modelParameters=optional_params, modelParameters=optional_params,
prompt=prompt, prompt=prompt,
completion=response_obj['choices'][0]['message'], completion=response_obj['choices'][0]['message'].json(),
usage=Usage( usage=Usage(
prompt_tokens=response_obj['usage']['prompt_tokens'], prompt_tokens=response_obj['usage']['prompt_tokens'],
completion_tokens=response_obj['usage']['completion_tokens'] completion_tokens=response_obj['usage']['completion_tokens']

16
litellm/utils.py
View file

@ -178,6 +178,14 @@ class Message(OpenAIObject):
# Allow dictionary-style assignment of attributes # Allow dictionary-style assignment of attributes
setattr(self, key, value) setattr(self, key, value)
def json(self, **kwargs):
try:
return self.model_dump() # noqa
except:
# if using pydantic v1
return self.dict()
class Delta(OpenAIObject): class Delta(OpenAIObject):
def __init__(self, content=None, role=None, **params): def __init__(self, content=None, role=None, **params):
super(Delta, self).__init__(**params) super(Delta, self).__init__(**params)
@ -817,16 +825,17 @@ class Logging:
) )
# print(f"original response in success handler: {self.model_call_details['original_response']}") # print(f"original response in success handler: {self.model_call_details['original_response']}")
try: try:
print_verbose(f"success callbacks: {litellm.success_callback}") print_verbose(f"success callbacks: {litellm.success_callback}")
## BUILD COMPLETE STREAMED RESPONSE ## BUILD COMPLETE STREAMED RESPONSE
complete_streaming_response = None complete_streaming_response = None
if self.stream == True and self.model_call_details.get("litellm_params", {}).get("acompletion", False) == True: if self.stream == True and self.model_call_details.get("litellm_params", {}).get("acompletion", False) == True:
# if it's acompletion == True, chunks are built/appended in async_success_handler # if it's acompletion == True, chunks are built/appended in async_success_handler
self.streaming_chunks.append(result)
if result.choices[0].finish_reason is not None: # if it's the last chunk if result.choices[0].finish_reason is not None: # if it's the last chunk
complete_streaming_response = litellm.stream_chunk_builder(self.streaming_chunks, messages=self.model_call_details.get("messages", None)) complete_streaming_response = litellm.stream_chunk_builder(self.streaming_chunks, messages=self.model_call_details.get("messages", None))
else: else:
# this is a completion() call # this is a completion() call
if self.stream: if self.stream == True:
print_verbose("success callback - assembling complete streaming response") print_verbose("success callback - assembling complete streaming response")
if result.choices[0].finish_reason is not None: # if it's the last chunk if result.choices[0].finish_reason is not None: # if it's the last chunk
print_verbose(f"success callback - Got the very Last chunk. Assembling {self.streaming_chunks}") print_verbose(f"success callback - Got the very Last chunk. Assembling {self.streaming_chunks}")
@ -5766,14 +5775,13 @@ class CustomStreamWrapper:
if processed_chunk is None: if processed_chunk is None:
continue continue
## LOGGING ## LOGGING
threading.Thread(target=self.logging_obj.success_handler, args=(processed_chunk,)).start() # log response
asyncio.create_task(self.logging_obj.async_success_handler(processed_chunk,)) asyncio.create_task(self.logging_obj.async_success_handler(processed_chunk,))
return processed_chunk return processed_chunk
raise StopAsyncIteration raise StopAsyncIteration
else: # temporary patch for non-aiohttp async calls else: # temporary patch for non-aiohttp async calls
# example - boto3 bedrock llms # example - boto3 bedrock llms
print_verbose(f"ENTERS __NEXT__ LOOP")
processed_chunk = next(self) processed_chunk = next(self)
print_verbose(f"PROCESSED CHUNK IN __ANEXT__: {processed_chunk}")
asyncio.create_task(self.logging_obj.async_success_handler(processed_chunk,)) asyncio.create_task(self.logging_obj.async_success_handler(processed_chunk,))
return processed_chunk return processed_chunk
except StopAsyncIteration: except StopAsyncIteration: