/leptonai

A Pythonic framework to simplify AI service building

Primary LanguagePythonApache License 2.0Apache-2.0

Lepton AI

A Pythonic framework to simplify AI service building

HomepageAPI PlaygroundExamplesDocumentationCLI ReferencesTwitterBlog

The LeptonAI Python library allows you to build an AI service from Python code with ease. Key features include:

  • A Pythonic abstraction Photon, allowing you to convert research and modeling code into a service with a few lines of code.
  • Simple abstractions to launch models like those on HuggingFace in few lines of code.
  • Prebuilt examples for common models such as Llama, SDXL, Whisper, and others.
  • AI tailored batteries included such as autobatching, background jobs, etc.
  • A client to automatically call your service like native Python functions.
  • Pythonic configuration specs to be readily shipped in a cloud environment.

Getting started with one-liner

Install the library with:

pip install -U leptonai

This installs the leptonai Python library, as well as the commandline interface lep. You can then launch a HuggingFace model, say gpt2, in one line of code:

lep photon run --name gpt2 --model hf:gpt2 --local

If you have access to the Llama2 model (apply for access here) and you have a reasonably sized GPU, you can launch it with:

# hint: you can also write `-n` and `-m` for short
lep photon run -n llama2 -m hf:meta-llama/Llama-2-7b-chat-hf --local

(Be sure to use the -hf version for Llama2, which is compatible with huggingface pipelines.)

You can then access the service with:

from leptonai.client import Client, local
c = Client(local(port=8080))
# Use the following to print the doc
print(c.run.__doc__)
print(c.run(inputs="I enjoy walking with my cute dog"))

Fully managed Llama2 models and CodeLlama models can be found in the playground.

Many standard HuggingFace pipelines are supported - find out more details in the documentation. Not all HuggingFace models are supported though, as many of them contain custom code and are not standard pipelines. If you find a popular model you would like to support, please open an issue or a PR.

Checking out more examples

You can find out more examples from the examples repository. For example, launch the Stable Diffusion XL model with:

git clone git@github.com:leptonai/examples.git
cd examples
lep photon run -n sdxl -m advanced/sdxl/sdxl.py --local

Once the service is running, you can access it with:

from leptonai.client import Client, local

c = Client(local(port=8080))

img_content = c.run(prompt="a cat launching rocket", seed=1234)
with open("cat.png", "wb") as fid:
    fid.write(img_content)

or access the mounted Gradio UI at http://localhost:8080/ui. Check the README file for more details.

A fully managed SDXL is hosted at https://dashboard.lepton.ai/playground/sdxl with API access.

Writing your own photons

Writing your own photon is simple: write a Python Photon class and decorate functions with @Photon.handler. As long as your input and output are JSON serializable, you are good to go. For example, the following code launches a simple echo service:

# my_photon.py
from leptonai.photon import Photon

class Echo(Photon):
    @Photon.handler
    def echo(self, inputs: str) -> str:
        """
        A simple example to return the original input.
        """
        return inputs

You can then launch the service with:

lep photon run -n echo -m my_photon.py --local

Then, you can use your service as follows:

from leptonai.client import Client, local

c = Client(local(port=8080))

# will print available paths
print(c.paths())
# will print the doc for c.echo. You can also use `c.echo?` in Jupyter.
print(c.echo.__doc__)
# will actually call echo.
c.echo(inputs="hello world")

For more details, checkout the documentation and the examples.

Contributing

Contributions and collaborations are welcome and highly appreciated. Please check out the contributor guide for how to get involved.

License

The Lepton AI Python library is released under the Apache 2.0 license.

Developer Note: early development of LeptonAI was in a separate mono-repo, which is why you may see commits from the leptonai/lepton repo. We intend to use this open source repo as the source of truth going forward.