Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
80 lines (65 loc) · 4.67 KB

CONTRIBUTING.md

File metadata and controls

80 lines (65 loc) · 4.67 KB

Contributing Guideline

Welcome to the PromptTrail project! We're glad you're considering contributing to our project!

Development Setup

  • We use rye to manage our Python development environment.
  • You can set up your development environment as follows:
    • Install rye (See their website for the latest installation instructions)
      • curl -sSf https://rye-up.com/get | bash
    • Clone this repository to your desired location
      • git clone https://github.com/combinatrix-ai/PromptTrail.git
    • Set up the environment
      • cd PromptTrail
      • Install dependencies
        • rye sync
      • Activate the environment
        • rye shell
    • Run an example
      • python -m examples.example_openai.py
        • You may need to set the OPENAI_API_KEY environment variable to run this example.

Before Submitting Pull Requests

  • Please ensure that your code passes all tests. (Adding tests is always welcome!)
    • rye run test
  • Please ensure that your code is auto-formatted. (black, autoflake, isort)
    • rye run format
  • Please ensure that your code passes the lint check. (mypy)
    • rye run lint
  • Please build the documentation if you change the public interface. (sphinx)
    • rye run doc

Note: You can run all of these commands at once using rye run all. GitHub Actions will automatically run these commands in check mode when you submit a pull request. If any of these checks fail, please run these commands locally and fix any errors.

License

  • This project is currently licensed under the Elastic License 2.0.

    • If you're comfortable with Elasticsearch, you should be comfortable with this project as well.
    • The Elastic License 2.0 is not OSI-compliant, but if you're okay with the Llama 2 License, you should be okay to use a non-OSI-compliant license.

  • We chose this license because we plan to provide a web service that allows everyone, including non-developers, to collaboratively create, edit, test, and host templates and agents online.

    • We aim to build a Huggingface/Github for LLM programming!
    • If you want to build this kind of service, we hope you'll contact us first if you want to use this project.

  • In other use cases, which means almost everything you might do, you can use this project commercially.

  • Therefore, before you contribute, you must agree that your contribution will be licensed under the Elastic License 2.0 or other licenses we have chosen for this project.

    • You grant us a worldwide, royalty-free, exclusive, perpetual, and irrevocable license, with the right to transfer an unlimited number of non-exclusive licenses or to grant sublicenses to third parties, under the Copyright covering the Contribution to use the Contribution by all means, including, but not limited to:
      • Publishing the Contribution
      • Modifying the Contribution
      • Preparing derivative works based upon or containing the Contribution and/or combining the Contribution with other Materials
      • Reproducing the Contribution in its original or modified form
      • Distributing, making the Contribution available to the public, displaying, and publicly performing the Contribution in its original or modified form

  • The sentences above may seem a little bit intimidating, but in essence, we want you to declare that your contribution can be used freely in this project so that other people can use it without worry.

  • Note: The Elastic License 2.0 is not compatible with GPL, so we may not use GPL libraries in this project.

Coding Principles

  • If you know what an LLM is, you must be able to use PromptTrail.
  • Agent (Flow) as Code
    • Agent that can be written in one place by code
      • Hook-based agent definition like PyTorch Lightning
  • Provide an easy way to debug prompt program
    • Record everything for later inspection
    • Easy to read error messages with template id, hook name, etc... is included
  • Intuitive and explicit (but sometimes convention)
    • Everything evolves fast here. You can't be sure what is right now. So explicit is better than implicit. Code is better than document.
      • No hidden templates and configurations
      • Every parameter should be passed explicitly and be able to be understood by types
        • Easy to work with on VSCode and JetBrains IDEs
    • Everything must be clear by class inheritance and types. I don't want to read docs.
      • Unified access to templates, parameters of agents
      • Hook-based agent definition
      • More default values

Type Checking

  • We use pylance (strict) in VSCode and mypy.
    • We prefer repeating duplicated type checking code over using Any or # type: ignore.