FunctionCalling: Automatically wrap FunctionCall in a Part when present in a Parts array

[williamito]

Description of the feature request:

When sending FunctionResponse in a multi-turn messages array, I have to wrap it in glm.Part manually:

s = Struct()
s.update({'result': result})
function_response = glm.Part(
    function_response=glm.FunctionResponse(name='named_function', response=s))

messages = [
    ...
    {'role':'user',
     'parts': [function_response]}
]

Ideally the SDK should automatically wrap with Part like other types, allowing for

{'role':'user',
     'parts': [glm.FunctionResponse(name='named_function', response=s)]
}

What problem are you trying to solve with this feature?

Simplify SDK usage

Any other information you'd like to share?

No response

[rebeatle]

Description of the feature request:

When sending FunctionResponse in a multi-turn messages array, I have to wrap it in glm.Part manually:

s = Struct()
s.update({'result': result})
function_response = glm.Part(
    function_response=glm.FunctionResponse(name='named_function', response=s))

messages = [
    ...
    {'role':'user',
     'parts': [function_response]}
]

Ideally the SDK should automatically wrap with Part like other types, allowing for

{'role':'user',
     'parts': [glm.FunctionResponse(name='named_function', response=s)]
}

What problem are you trying to solve with this feature?

Simplify SDK usage

Any other information you'd like to share?

No response

Not sure if they fixed this but i tried your solution but didnt work. Im using a JSON chat history to record everthing and that format is not acceptable. I am using this format just in case someone wants to try something different : First i am looking for function call on the response like this :

            for part in response.candidates[0].content.parts:
                if part.function_call and part.function_call.name:

Then im passing "part" to build "part.function_call.name" finally im doing the append to "historial" to save it later to a JSON .

        function_response = {
            "role": "tool",
            "parts": [
                {
                    "function_response": {
                        "name": part.function_call.name,
                        "response": {"result": resultado}
                    }
                }
            ]
        }
historial.append(function_response)

With this you can use

response = modelo.generate_content(
                historial)

directly and avoid using the " Low level access" code . and keep it simple.