ChatCompletion does not handle functions = None

Describe the bug

Firstly I noticed how you have integrated the chat_completion_request function into the Python OpenAI package instead of writing it out in code. I have used prior version and the new version but am finding the newer integrated version giving me issues.

Specificaly it seems that it does not handle the case where functions = None very well. There does not appear to be a way to tell 'ChatCompletion' to not run any functions other than giving it a fake function and then setting the function_call flag to 'none'.

Example error:
error_code=None error_message="None is not of type 'array' - 'functions'" error_param=None error_type=invalid_request_error message='OpenAI API error received' stream_error=False

It would be good if it could proces requests when functions is set to none and function_call is set to 'none'.

To Reproduce

Pass in functions=Non and function_call='none' to the code below.

Code snippets

def chat_completion_request(messages, functions=None, function_call="auto", model=GPT_MODEL):
    if functions is None: 
        function_call ="none"
   
    r= openai.ChatCompletion.create(
        model=GPT_MODEL,
        messages=messages,
        functions=functions,
        function_call=function_call,  # auto is default, but we'll be explicit
    )
    return r

OS

Windows 11

Python version

Latest

Library version

Latest

Realised I was forgetting to return the result with a .json() at the end.

It looks like if you dig all the way to the API request, the params (i.e., all keyword arguments except api_key, api_base, api_type, request_id, api_version and organization) dictionary is serialized to json here: https://github.com/openai/openai-python/blob/b82a3f7e4c462a8a10fa445193301a3cefef9a4a/openai/api_requestor.py#L553C37-L553C37

That's probably going to result in different behavior if you specify functions=None vs specifying nothing at all.

This is "fixed" in the v1 release as we've moved to an explicit None vs omitted model where we have an internal sentinel value to signify if an argument was given or not.

If you try to pass functions=None you'll get a static type error and the API will raise an error at runtime.

Start and finish streaming trace in impl metod (openai#540)

Navigation Menu

Repository navigation

ChatCompletion does not handle functions = None #538

Describe the bug

To Reproduce

Code snippets

OS

Python version

Library version

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

ChatCompletion does not handle functions = None #538

Description

Describe the bug

To Reproduce

Code snippets

OS

Python version

Library version

Activity

badrul1 commented on Jul 13, 2023

jeremy-fisher-nd commented on Jul 25, 2023

RobertCraigie commented on Nov 6, 2023

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions