How To

This section of the documentation contains common code snippets for common tasks with Ghostfunctions.

Use a different OpenAI Model

By default, AI Ghostfunctions will use OpenAI’s gpt-4 model if the OpenAI API Key has access to it, optionally falling back to gpt-3.5-turbo if the API request fails with a openai.InvalidRequestError. You can configure the model to use gpt-3.5-turbo by default by using the ai_callable parameter on the @ghostfunction decorator:

>>> from ai_ghostfunctions import ghostfunction
>>> import openai
>>> def query_openai_35(**kwargs):
>>>     """Query OpenAI API with `query` and return the response."""
>>>     return openai.ChatCompletion.create(
>>>         model='gpt-3.5-turbo',
>>>         **kwargs
>>>     )
>>>
>>> @ghostfunction(ai_callable=query_openai_35)
>>> def complete_the_rhyme(phrase: str) -> str:
>>>     """Output a word that completes phrase with a rhyme."""
>>>
>>> complete_the_rhyme(phrase="If you blow a double bubble, you're sure to be in")
'trouble'

Ensure deterministic results

One drawback of using ghostfunctions is that the same inputs to a ghostfunction may return different outputs, even when temperature is set to 0. To keep the library small and focused, and to give users flexibility in caching configurations, I recommend using a caching decorator to ensure determinism. One simple, built-in caching mechanism is functools.lru_cache:

>>> from functools import lru_cache
>>> from ai_ghostfunctions import ghostfunction
>>>
>>> @lru_cache
>>> @ghostfunction
>>> def chaos(seed: int) -> str:
>>>     """Returns chaos!!!"""
>>>
>>> chaos(seed=1)
'chaos!!!'
>>> chaos(seed=2)
'Chaos!!!'
>>> chaos(seed=1)
'chaos!!!'
>>> chaos(seed=2)
'Chaos!!!'

The cachetools third-party library also contains other implementations of caching decorators.

Choose from several different results from OpenAI API

OpenAI exposes a parameter into their ChatCompletion API to return multiple different responses from the model in a single API call. Ghostfunctions exposes an API to return several of these choices from OpenAI, and aggregate them into a final result:

>>> from ai_ghostfunctions import ghostfunction
>>> import openai
>>> def query_openai_35(**kwargs):
>>>     """Query OpenAI API with `query` and return the response."""
>>>     return openai.ChatCompletion.create(
>>>         model='gpt-3.5-turbo',
>>>         n=4,
>>>         temperature=.6,
>>>         **kwargs
>>>     )
>>>
>>> @ghostfunction(
>>>     ai_callable=query_openai_35,
>>>     aggregation_function=",".join
>>> )
>>> def word_that_starts_with(letter: str = "n") -> str:
>>>     """Output a word that starts with `letter`."""
>>>
>>> word_that_starts_with('n')
'nose,nose,nose,nose,nose'

You can use this functionality to vet responses from the AI.