When tried with gpt-3.5-turbo-16k Cannot Use Function Calling

[yigit353]

The project works excellently with gpt-4-turbo, although with some hiccups.

When I change the model to gpt-3.5-turbo-16k, I get the error below. Is that a known issue of the AutoGEN, or is there something I should change too?

***** Response from calling function "data_retreiver.getCode" *****
Error: Function data_retreiver.getCode not found.

The error is caused by this

chat_manager (to chat_manager):

Show me the cheapest flights from Istanbul Airport to Amsterdam, leaving on the
    21st of December 2023 and returning on the 23rd of December 2023.

--------------------------------------------------------------------------------
data_retriever (to chat_manager):

***** Suggested function Call: data_retreiver.getCode *****
Arguments: 
{
  "originLocationCode": "IST",
  "destinationLocationCode": "AMS",
  "departureDate": "2023-12-21",
  "returnDate": "2023-12-23",
  "adults": 1,
  "maxPrice": 0,
  "max": 10
}
***********************************************************

--------------------------------------------------------------------------------
user_proxy (to chat_manager):

***** Response from calling function "data_retreiver.getCode" *****
Error: Function data_retreiver.getCode not found.
*******************************************************************

...

BadRequestError: Error code: 400 - {'error': {'message': "'data_retreiver.getCode' does not match '^[a-zA-Z0-9_-]{1,64}$' - 'messages.2.function_call.name'", 'type': 'invalid_request_error', 'param': None, 'code': None}}

gpt-4-turbo generates the correct function call while gpt-3.5-turbo makes up inexisting function:

***** Suggested function Call: get_flight_data *****

[john-adeojo]

Hmm, to be honest I'm not completely sure where this error is coming from. Did you change the model in the configuration file too?

[yigit353]

configurations.json

[
        {
            "model": "gpt-3.5-turbo-16k-0613",
            "api_key": "sk-xxxxxxxxxxxxxx"
        } ,
        {
            "model": "gpt-4-1106-preview",
            "api_key": "sk-yyyyyyyyyyyyyyyy"
        } 
    ]

I even used different API keys for both models. And in the notebook:

config_list = autogen.config_list_from_json(
    env_or_file="configurations.json",
    file_location=configurations_path,
    filter_dict={
        # "model": ["gpt-4-1106-preview"],
        "model": ["gpt-3.5-turbo-16k-0613"]
    },
)

It makes a call to gpt3.5-turbo-16k, I can confirm it from the usage:

However, even though I changed the seed and temperature for the data retriever agent, it made up the same wrong function calls every time.

llm_config_data_retriever = {
    "functions":get_flight_data_functions,
    "config_list": config_list,
    "seed": 150,
    "temperature": 0.2
}