shuffle-sdk 0.0.30

The SDK used for Shuffle

Shuffle SDK

This is the SDK used for app development, testing and production of ALL apps in Shuffle. Works with manual runs, Docker, k8s, cloud serverless.

Released under Python pip for usage outside of Shuffle

Python apps: https://github.com/shuffle/python-apps All apps: https://shuffler.io/search

Usage with Shuffle

Refer to the Shuffle App Creation docs

Build

docker build . -t shuffle/shuffle:app_sdk

Download

pip install shuffle_sdk

Usage

from shuffle_sdk import AppBase

class Example(AppBase):
    def sample_function(self, paramname):
        return f"Hello {paramname}"

if __name__ == "__main__":
    Example.run()

Testing Shuffle Apps

With the above function as an example

python3 app.py --standalone --action=sample_function paramname=World

NOT REQUIRED: These are reserved parameters, allowing you to set variables for testing such as local backends to upload into Shuffle during tests:

--shuffle_url="http://custom_backend_url"
--shuffle_authorization="auth"
--shuffle_executionid="execution_id"

Example wit Shuffle Tools+Liquid and the Shuffle Tools app and the "repeat back to me" function

python3 app.py --standalone --action=repeat_back_to_me '--call={{ "hello" | replace: "o", "lol" }}'

Example using Shuffle Tools Shuffle actions within the "execute_python" function to get emails from Outlook (app.py)

python3 app.py --standalone --action=execute_python 'code=print(shuffle.run_app(app_id="accdaaf2eeba6a6ed43b2efc0112032d", action="get_emails"))'

Example LLM inference with the Shuffle-AI app. Supports GPU and requires ollama installed and serving.

python3 app.py --standalone --action=run_llm 'input=convert the following data into a python list of valid ips: 12.3.4.4'

If successful, the output of the function will show in your CLI.

Testing functions inside Shuffle App images manually

This is mostly the same, but paths and docker is a part of the command.

docker run frikky/shuffle:shuffle-ai_1.0.0 python3 /app/app.py standalone --action=run_llm '--input=llm, please answer this question thanks'

With GPU's active in Docker for the LLM (requires Nvidia Container Toolkit):

docker run --gpus all GPU_LAYERS=8 frikky/shuffle:shuffle-ai_1.0.0 python3 /app/app.py standalone --action=run_llm '--input=llm, please answer this question thanks'

Building a fully functional Shuffle App

Look at the documentation on our website

Adding new Liquid filters

Add a function along these lines:

@shuffle_filters.register
def md5(a):
    a = str(a)
    return hashlib.md5(a.encode('utf-8')).hexdigest()

This can be used as {{ "string" | md5 }}, where "string" -> the a parameter of the function

(Scale testing) Running the Shuffle Tools within the Shuffle Tools app locally

1. Set your env SHUFFLE_APP_EXPOSED_PORT=8080 and SHUFFLE_SWARM_CONFIG=run
2. Run the Shuffle Tools app as a webserver: python3 app.py
3. Send a local request to the Shuffle Tools app to run an action within an action. Make sure to add an authorization & execution_id that exists

curl -XPOST http://localhost:8080/api/v1/run -H "Content-Type: application/json" -d '{
    "action": {"app_name": "shuffle tools", "name": "execute_python", "parameters": [{
        "name": "code", 
        "value": "singul.config[\"url\"] = \"https://shuffler.io\";print(singul.run_app(app_id=\"3e2bdf9d5069fe3f4746c29d68785a6a\", action=\"repeat_back_to_me\", parameters=[{\"name\": \"call\", \"value\": \"testing\"}]))"}]
    }, 
    "authorization": "",
    "execution_id": ""
}
'

You can replace the singul.config URL with a local instance for local testing as well.