Getting Started with Keymaker

Keymaker is a Python library that provides a powerful and flexible way to control the output of large language models. This guide will walk you through installing Keymaker and using it to generate some completions including:

with constraints
different models
streaming

Installation

Install Keymaker with:

pip install headjack-keymaker

For HuggingFace support:

pip install headjack-keymaker[huggingface]

For LlamaCpp support:

pip install headjack-keymaker[llamacpp]

For everything:

pip install headjack-keymaker[all]

Basic Usage

Import Keymaker:

from keymaker import Prompt

Create a prompt:

prompt = Prompt("Dogs are known for being man's")

Import a model - here we will use Huggingface, but Keymaker supports OpenAI chat completion, OpenAI completion, LlamaCPP, and OpenAI API compatible models.

from keymaker.models import Huggingface

model = Huggingface("gpt2")

Make a single one-off completion of the prompt with a model:

completed = await prompt.complete(model=model, max_tokens=3)

The completed prompt text is now available:

print(completed)
# Dogs are known for being man's best friend.

Formatting and Constraints

Beyond one-off completions, Keymaker supports an extremely flexible formatted completion and of course constraints.

Import a constraint and CompletionConfig:

from keymaker import CompletionConfig, Prompt
from keymaker.models import Huggingface, chatgpt
from keymaker.constraints import RegexConstraint, OptionsConstraint

Set up the model:

model = Huggingface("gpt2")

Define a function to print the stream:

async def print_stream(completion):
    if completion:
        print(completion)

Provide a default model and stream which will be used if we do not specify otherwise:

prompt = Prompt(
    "My name is {} and I am {age} years old. I like {hobbies}",
    model, 
    stream=print_stream,
)

Define possible hobbies:

possible_hobbies = {"Reading", "Sports", "Gaming", "Traveling", "Cooking"}

Set the OpenAI API key for using chatgpt:

import openai

openai.api_key = ""

chat_model = chatgpt()

Here, we define a function to generate unique hobbies:

def hobbies(prompt: Prompt):
    """Generate Unique Hobbies"""
    while True:
        if prompt.completions["hobbies"] is None:
            hobbies_selected = set()
        elif isinstance(prompt.completions.hobbies, list):
            yield ", "
            hobbies_selected = set(prompt.completions.hobbies)
        else:
            yield ", "
            hobbies_selected = {prompt.completions.hobbies}
        if len(hobbies_selected) > 1:
            yield "and Soccer."
            break
        hobbies_left = possible_hobbies - hobbies_selected
        yield CompletionConfig(
            model=chat_model, constraint=OptionsConstraint(hobbies_left)
        )

Keymaker supports the following as format parameters:

Stringable: Anything that can be cast to a str.
CompletionConfig: A single parameterized Completion.
Callable[[Prompt], Stringable | CompletionConfig]: Dynamically change the prompt completion by returning a SINGLE static Stringable or request a paramterized Completion with a function that takes the current prompt at that point of completion.
Callable[[Prompt], Generator[Stringable | CompletionConfig]]: Dynamically change the prompt completion by returning as many static Stringable or paramterized Completions with a Generator

Fill the prompt:

filled = await prompt.format(
    "Alice",
    age=CompletionConfig(
        constraint=RegexConstraint(r"[1-9]{1,2}", terminate_on_match=True), map_fn=int
    ),
    hobbies=hobbies,
)

The streamed output will be:

My name is 
Alice
 and I am 
33
 years old. I like 
Reading
, 
Travel
, 
and Soccer.

You can access the age completion:

age = filled.completions.age

Finally, print the filled prompt and the age:

print(filled)
print(age, type(age.value))
# My name is Alice and I am 33 years old. I like Reading, Traveling, and Soccer.
# 33 <class 'int'>

Congratulations! You’ve completed the guide to getting started with Keymaker.

Docs

Getting Started

Getting Started with Keymaker

Installation

Basic Usage

Formatting and Constraints