🌱 Sprinkler 🌱

🛠️ The Tool for constructing pipeline for your application

Origin from graudation project from Sungkyunkwan University

☄️ How to install (TBD)

Install easily with PyPI - Sprinkler !!!

pip install sprinkler

🦷 Requirements

Please install below packages for Sprinkler. We recommed using pip command as sprinkler.

• python 3.8+
• pydantic 2.4.2+
• openapi 0.28.0 (optional - for chat completion of GPT)
• pygraphviz 1.11 (optional - for visualization of pipeline)

If you want to visualize your own pipeline, after install pygraphviz packages, you sholuld install Graphviz and C/C++ Compiler depeding on your OS. For Graphviz with Linux Ubuntu, installation command is below.

sudo apt-get install graphviz graphviz-dev

If you need some installation guide of Graphviz on different OS, The instruction manual is here

😆 What is Sprinkler for?

As apperance of ChatGPT, LLM (Large Language Model) becomes the talk of all programming parts. As active research on LLM is ongoing, there are many efforts to utilize LLM to application for user. Although LLM is so powerful itself, the limit is also clear. So, we should divide the whole task we want solve utilizing LLM into smaller tasks and constructe them into pipeline sequentialy or in parallel.

Sprinkler is developed for this purposes. From Sprinkler, you can imporve reusability and stability. Sprinkler has main three components which are runnable.

• Task - the runnalbe which operates one function
• Pipeline - the runnable which operates runnables sequentially
• Group - the runnable which operates runnables in parallel

You can make your own pipeline system with these components. For more information, Let's see next section for usage.

👀 Examples

The one basic examples of just repeating something using Task, Pipeline, Group, Context is like below.

from concurrent.futures import ProcessPoolExecutor

from sprinkler import Pipeline, Group, Task, Ctx


@Task('camel')
def make_camel(text: str) -> str:
    return ''.join(
        word.lower() if i == 0 else word.capitalize()
        for i, word in enumerate(text.split())
    )

def repeat(text: str, times: Ctx[int]) -> str:
    return ''.join(ch * times for ch in text)

def tile(text: str, times: Ctx[int]) -> str:
    return text * times

pipeline = Pipeline('pipeline', context={'times': 2}).add(
    make_camel,
    Group('group').add(
        Task('repeat', repeat),
        Task('tile', tile)
    )
)

with ProcessPoolExecutor() as executor:
    pipeline.run('python sprinkler', __executor__=executor)

# output becomes 
# {
#   'repeat': 'ppyytthhoonnSSpprriinnkklleerr',
#   'tile': 'pythonSprinklerpythonSprinkler'
# }

With GPT Chatcompletion, you can construct your pipeline like below.

from sprinkler.prompt_template import PromptTemplate, SystemPromptTemplate

from sprinkler.runnable.task import PromptTask
from sprinkler.runnable.task import ChatCompletionTask


messages = [SystemPromptTemplate('You are a fan of baseball team named {team}'),
            PromptTemplate('Have you bought {team} merchandise?')]

task_prompt = PromptTask('prompt',{'messages': messages})
task_chat = ChatCompletionTask('chat')

pipeline = Pipeline('pipeline').add(task_prompt, task_chat)

output = pipeline.run({'team': 'sprinkler'})

# output becomes like (example)
# Yes, as a fan of the Sprinkler baseball team, I have bought various sprinkler merchandise. This includes hats, jerseys, t-shirts, keychains, and even a mini sprinkler for my garden. Showcasing my support for the team and representing them through merchandise is a fun way to connect with fellow fans and show my love for the Sprinkler baseball team

🧑🏻‍💻 Contributing

Open Source projects and cultures are actively developing in this era. Our project is also open to every contributors and feedback from users.