Deserve

Deserve is a nanoframework for serving ML models. Flasker than Flask, faster than FastAPI, Deserve is asynchronous, lightweight and simple.

Features

• 🤙 Remote procedure call (RPC) architecture. There are no endpoints, methods, paths, nor resources to make decisions about — just the host:port.
• 📦 Send JSON, receive JSON, client-side. Accept a Python object, return an object, server-side. Conversions happen under the hood.

Installing

$ pip install deserve

Also install an ASGI server such as Uvicorn or Hypercorn.

$ pip install hypercorn

Quickstart

This example uses the 🤗 Transformers library.

# Save this as example.py
import deserve
from transformers import pipeline

# Load your model
classifier = pipeline('sentiment-analysis')

@deserve
async def predict(payload: object) -> object:
    return classifier(payload)

Run the server using the names of your file (example.py) and function (predict).

$ hypercorn example:predict

[INFO] Running on http://127.0.0.1:8000

Get some predictions.

$ curl localhost:8000 --data '["This is the simplest framework.", "You deserve it!"]'

[{"label": "POSITIVE", "score": 0.799}, {"label": "POSITIVE", "score": 0.998}]

Serialization and deserialization

Deserve takes care of converting the request and response payloads based on the content-type and accept headers of the request.

List of supported content-types and the deserialized payload that is passed to the inference handler.

List of supported accept headers and the serialized payload that is returned.