Implement chains and use them for OpenAI functions

[simonw]

I'm going to handle chaining as a separate, post-0.5 issue.

Originally posted by @simonw in https://github.com/simonw/llm/issues/85#issuecomment-1630178102

[simonw]

Here's an experimental .chain() method on Model which I'm about to remove - I think this should happen on Conversation from #85 instead.

https://github.com/simonw/llm/blob/b38b8314b980cbf4b1d2034809faffc2437ed608/llm/models.py#L217-L243

[simonw]

My latest thinking on this is that it can be part of the conversations mechanism. I'm thinking something like this:

def search_wikipedia(q: str) -> str:
    "Search for wikipedia articles matching q and return their summary"
    ...

model = llm.get_model("4")

conversation = model.conversation(functions=[search_wikipedia])
chain_response = conversation.chain("Tell me about Agatha Christie")
print(chain_response.last.text())

So .conversation() on the OpenAI Chat model class takes an extra functions= argument.

And .chain() takes a prompt and executes multiple prompts as part of handling those functions.

I'm very undecided about the .chain() mechanism so far though.

[simonw]

This is what a functions call partially looks like, from https://til.simonwillison.net/gpt3/openai-python-functions-data-extraction

completion = openai.ChatCompletion.create(
    model="gpt-3.5-turbo-0613",
    messages=[{"role": "user", "content": "I went to London and then stopped in Istanbul and Utrecht."}],
    functions=[
        {
            "name": "extract_locations",
            "description": "Extract all locations mentioned in the text",
            "parameters": {
                "type": "object",
                "properties": {
                    "locations": {
                        "type": "array",
                        "items": {
                            "type": "object",
                            "properties": {
                                "name": {
                                    "type": "string",
                                    "description": "The name of the location"
                                },
                                "country_iso_alpha2": {
                                    "type": "string",
                                    "description": "The ISO alpha-2 code of the country where the location is situated"
                                }
                            },
                            "required": ["name", "country_iso_alpha2"]
                        }
                    }
                },
                "required": ["locations"],
            },
        },
    ],
    function_call={"name": "extract_locations"}
)
choice = completion.choices[0]
encoded_data = choice.message.function_call.arguments
print(json.dumps(json.loads(encoded_data), indent=4))

Here's the code where that would need to happen, currently buried deep in the .execute() method of the OpenAI default plugin: https://github.com/simonw/llm/blob/cb41409e2b499b76ec37db178c3b0fc042d59129/llm/default_plugins/openai_models.py#L205-L227

[simonw]

I think functions= and function_call= become optional parameters on the Chat.execute(...) method there. Then they get called by a custom Conversation.chain() method.