#well as a side note it is to call
wet shale
You have a description for a function
in the example
function_call = ai_response["choices"][0]["message"]["function_call"]
function_name = function_call["name"]
arguments = function_call["arguments"]
if function_name == "function_1":
one = eval(arguments).get("one")
print(f"Function 1 {one}")
return
if function_name == "function_2":
print('Function 2')
one = eval(arguments).get("one")
two = eval(arguments).get("two")
print(f"Function 2 {one}{two}")
return
else:
return```
This is where it would take the definition and pass the information to the actual functions
golden arch
You mean locally defined functions for after the callback?
wet shale
all this does
{
"name": "validate_haiku",
"description": "Validate the correctness of an AI generated haiku.",
"parameters": {
"type": "object",
"properties": {
"text": {
"type": "string",
"description": "The haiku text, with newline characters after lines."
}
},
"required": ["text"]
}
},
];```
is put together the variables to pass to the next stepthe API has no idea if the function exists or not beyond that though.
wet shale
No. the api has no access to your code and only knows what you pass to it
golden arch
Right, and you only pass it the function definition (functionDefs).
wet shale
Right, which is why you need to make the next step that handles it's response
golden arch
I know, but there's no point in doing that if I can't get the API call to work. Handling it afterwards will be trivial.
tidal bison
I just tried your code in JS and it works for me Justin, but I am using gpt-3.5-turbo-0613
your code is wrapped in an async function right?
golden arch
Yeah. It is. Let me try that model instead.
tidal bison
I don't have access to GPT-4 so I can't try the model you were using
golden arch
lol, that's so weird. It's working with 3.5-turbo, but not gpt-4.
wet shale
hAHahhah if thats the whole reason it's not trriggering
golden arch
the weather example works with either.
wet shale
You probably don't have access to gpt 4 /facepalm
golden arch
So it seems like gpt-4 works worse for function calling in at least this particular instance.
No, i do have access.
I regularly use gpt-4 and the weather example works for function calling with gpt-4.
tidal bison
Might be something worth reporting then 😄