lightinfer 0.1.0

A lightweight bridge to serve synchronous model inference via async FastAPI.

LightInfer

LightInfer is a lightweight, high-performance bridge for serving synchronous model inference code (PyTorch, TensorFlow, etc.) via an asynchronous FastAPI server.

It solves the "Blocking Loop" problem by efficiently isolating heavy computation in dedicated worker threads while maintaining a fully asynchronous, high-concurrency web frontend.

Features

• Zero-Blocking Architecture: Async Web Frontend + Sync Worker Threads.
• Efficient Bridge: Uses AsyncResponseBridge for zero-thread-overhead waiting.
• Streaming Support:
  • Native Server-Sent Events (SSE) for text streaming.
  • Binary Streaming for audio/video generation (with chunk buffering).
• Easy Integration: Wrap any Python class with an infer method.
• Context Isolation: Each worker runs in its own thread, ensuring safety for libraries like PyTorch.

Installation

pip install lightinfer

Quick Start

1. Define your Model

LightInfer wraps any class with an infer method. The arguments to infer are automatically mapped from the JSON request.

import time

class MyModel:
    def infer(self, prompt: str = "world"):
        # Simulate heavy work
        time.sleep(1)
        return {"message": f"Hello, {prompt}!"}

2. Start the Server

from lightinfer.server import LightServer

# Create your model instance
model = MyModel()

# Start server (you can pass a list of models to run multiple worker threads)
server = LightServer([model])
server.start(port=8000)

3. Make Requests

Standard Request:

import requests

# 'args' in JSON maps to positional arguments of infer()
# 'kwargs' in JSON maps to keyword arguments of infer()
resp = requests.post("http://localhost:8000/api/v1/infer", 
                     json={"args": ["LightInfer"]})
print(resp.json())
# Output: {'message': 'Hello, LightInfer!'}

Streaming Request:

If your model returns a generator, you can use streaming:

class StreamingModel:
    def infer(self, prompt: str):
        yield "Part 1"
        time.sleep(0.5)
        yield "Part 2"

Client side:

resp = requests.post("http://localhost:8000/api/v1/infer", 
                     json={"args": ["test"], "stream": True}, stream=True)

for line in resp.iter_lines():
    if line:
        print(line.decode('utf-8'))

Examples

Check the examples/ directory for ready-to-run scenarios:

• Simple LLM: Text-to-Text generation with SSE streaming.
• Streaming TTS: Text-to-Audio generation with binary chunk streaming.

CLI Usage

You can serve any model class directly from the terminal.

Format: lightinfer <module>:<Class>

Given a file my_model.py:

class MyModel:
    def infer(self, prompt: str):
        return f"Echo: {prompt}"

Run:

lightinfer my_model:MyModel --port 8000 --workers 2

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT