This is a collection of simple helper functions I use for developing language projects. Probably should be multiple independent repos.

CLI

Getting started

Copy the .env template and update it with your openai api key.

cp .env.template .env

Install the CLI.

pip install -e .

Write a prompt. Prompt files are simple text files that identify variables with the syntax [[variable]]. This is better than the basic python variable definition because for a lot of language projects it's useful to return the response as json and so the {variable} syntax can cause formatting issues.

Once you've written a prompt run this command:

llm_helper run path/to/prompt/file

This will then ask you for the variables in the prompt one at a time and afterwords will run the prompt against the OpenAI api. After it's run it stores all data about the completion in the folder:

~/.llm_helper_data

Future runs using run or chat will use the variables from your most recent run unless you would like to run it with different values.

You can review old runs using some of the CLI commands:

# Lists the last N runs you performed
llm_helper list [number_of_files]

# Prints out a more detailed list of the last runs
llm_helper tail [number_of_files]

llm_helper chat is the same structure as run with the exception that after the run is done it enters into a chat experience that you can then reask new questions

State

This is a simple base class that you can use to declare a set of language jobs that gives you the same printing functionality out of the box. The basic idea is you inherit from State, the state has a single status that is updated after every status method is executed and has a single state property that you can manipulate and update with your status methods. The job finishes when the status is end. It's easier to see a toy example:

from llm_helper.llm import render_and_run_prompt
from llm_helper.state import State

class TestState(State):
    def __init__(self):
        super().__init__(
            initial_status='birthday',
            status_map={
                'birthday': self.birthday,
                'end': self.end # you always need to declare this to make it work
            },
            initial_state={'received_birthday': False}
        )
    
    def birthday(self):
        render_and_run_prompt("prompts/test.prompt", variables={'individual': 'Harrison Ford'})
        self.state['received_birthday'] = True
        self.update_status('end')
    
if __name__ == '__main__':
    test_state = TestState()
    test_state.run()

The State object logs status history in the status_history property and the state history in the state_history property for you by default.