kani 0.3.4

kani (カニ) is a lightweight and highly hackable framework for chat-based language models with tool usage/function calling.

kani (カニ)

kani (カニ) is a lightweight and highly hackable framework for chat-based language models with tool usage/function calling.

Compared to other LM frameworks, kani is less opinionated and offers more fine-grained customizability over the parts of the control flow that matter, making it the perfect choice for NLP researchers, hobbyists, and developers alike.

kani comes with support for OpenAI models and LLaMA v2 out of the box, with a model-agnostic framework to add support for many more.

Interested in contributing? Check out our guide.

Read the docs on ReadTheDocs!

Read our preprint on arXiv!

Features

• Lightweight and high-level - kani implements common boilerplate to interface with language models without forcing you to use opinionated prompt frameworks or complex library-specific tooling.
• Model agnostic - kani provides a simple interface to implement: token counting and completion generation. Implement these two, and kani can run with any language model.
• Automatic chat memory management - Allow chat sessions to flow without worrying about managing the number of tokens in the history - kani takes care of it.
• Function calling with model feedback and retry - Give models access to functions in just one line of code. kani elegantly provides feedback about hallucinated parameters and errors and allows the model to retry calls.
• You control the prompts - There are no hidden prompt hacks. We will never decide for you how to format your own data, unlike other popular language model libraries.
• Fast to iterate and intuitive to learn - With kani, you only write Python - we handle the rest.
• Asynchronous design from the start - kani can scale to run multiple chat sessions in parallel easily, without having to manage multiple processes or programs.

Quickstart

kani requires Python 3.10 or above.

First, install the library. In this quickstart, we'll use the OpenAI engine, though kani is model-agnostic.

$ pip install "kani[openai]"

Then, let's use kani to create a simple chatbot using ChatGPT as a backend.

# import the library
from kani import Kani, chat_in_terminal
from kani.engines.openai import OpenAIEngine

# Replace this with your OpenAI API key: https://platform.openai.com/account/api-keys
api_key = "sk-..."

# kani uses an Engine to interact with the language model. You can specify other model 
# parameters here, like temperature=0.7.
engine = OpenAIEngine(api_key, model="gpt-3.5-turbo")

# The kani manages the chat state, prompting, and function calling. Here, we only give 
# it the engine to call ChatGPT, but you can specify other parameters like 
# system_prompt="You are..." here.
ai = Kani(engine)

# kani comes with a utility to interact with a kani through your terminal! Check out 
# the docs for how to use kani programmatically.
chat_in_terminal(ai)

kani makes the time to set up a working chat model short, while offering the programmer deep customizability over every prompt, function call, and even the underlying language model.

Function Calling

Function calling gives language models the ability to choose when to call a function you provide based off its documentation.

With kani, you can write functions in Python and expose them to the model with just one line of code: the @ai_function decorator.

# import the library
from typing import Annotated
from kani import AIParam, Kani, ai_function, chat_in_terminal
from kani.engines.openai import OpenAIEngine

# set up the engine as above
api_key = "sk-..."
engine = OpenAIEngine(api_key, model="gpt-3.5-turbo")


# subclass Kani to add AI functions
class MyKani(Kani):
    # Adding the annotation to a method exposes it to the AI
    @ai_function()
    def get_weather(
        self,
        # and you can provide extra documentation about specific parameters
        location: Annotated[str, AIParam(desc="The city and state, e.g. San Francisco, CA")],
    ):
        """Get the current weather in a given location."""
        # In this example, we mock the return, but you could call a real weather API
        return f"Weather in {location}: Sunny, 72 degrees fahrenheit."


ai = MyKani(engine)
chat_in_terminal(ai)

kani guarantees that function calls are valid by the time they reach your methods while allowing you to focus on writing code. For more information, check out the function calling docs.

Why kani?

Existing frameworks for language models like langchain and simpleaichat are opinionated and/or heavyweight - they edit developers' prompts under the hood, are challenging to learn, and are difficult to customize without adding a lot of high-maintenance bloat to your codebase.

We built kani to be more flexible, simple, and robust. kani is appropriate for everyone from academic researchers to industry professionals to hobbyists to use without worrying about under-the-hood hacks.

Docs

To learn more about how to customize kani with your own prompt wrappers, function calling, and more, read the docs!

Or take a look at the hands-on examples in this repo.

Demo

Want to see kani in action? Using 4-bit quantization to shrink the model, we run LLaMA v2 as part of our test suite right on GitHub Actions:

https://github.com/zhudotexe/kani/actions/workflows/pytest.yml?query=branch%3Amain+is%3Asuccess

Simply click on the latest build to see LLaMA's output!

Kani in the News

We are really excited and grateful to see people talking about Kani online. We are also trending on Papers With Code, GitHub, and OSS Insight. Check out some recent articles and videos below!

• Researchers from the University of Pennsylvania Introduce Kani: A Lightweight, Flexible, and Model-Agnostic Open-Source AI Framework for Building Language Model Applications
• Unlocking AI Potential: Unveiling Kani, the Groundbreaking Open-Source Framework Revolutionizing Large Language Model Applications
• Kani: A Lightweight and Customizable Framework for Language Model Applications
• Introducing Kani (Sanskrit Word): A Game-Changing Open-Source AI Framework for Language Models
  • Kani was originally named after the Japanese word for crab and coincidentally means "knowledge" in Sanskrit.
• kani: lightweight LLM framework (Japanese)
• Top Trending LLM Projects of the Week: Dive into the Future of Tech! 🚀

Citation

If you use Kani, please cite us as:

@misc{zhu2023kani,
      title={Kani: A Lightweight and Highly Hackable Framework for Building Language Model Applications}, 
      author={Andrew Zhu and Liam Dugan and Alyssa Hwang and Chris Callison-Burch},
      year={2023},
      eprint={2309.05542},
      archivePrefix={arXiv},
      primaryClass={cs.SE}
}