Modelzilla

This library turns any model class into a CLI executable. If you're tired of writing a lot of boilerplate code to run your model, this library is for you. It is a lightweight Python package that enables developers to transform any AI model into a fully functional Command-Line Interface (CLI) plugin.

Key features:

• Automatic CLI Generation: Automatically generate CLI commands for models
• Customizable Commands: Each model will have its own CLI parameters
• Seamless Integration: Just inherit from the CLIPlugin class and you're ready to go

Installation

Pip install the modelzilla package in a Python>=3.10 environment.

pip install modelzilla

Quickstart

1. Turn any model into a CLI executable

Let's say you have a model class that you want to turn into a CLI executable. You can do this by inheriting from the CLIPlugin class and implementing the inference method.

import supervision as sv

from modelzilla.plugins import CLIPlugin # Import the CLIPlugin class

class MyModel(CLIPlugin):
    def __init__(self, model_path: str): # Add any parameters you want to the CLI
        self.model = load(model_path) # Load your model

    def inference(self, image) -> sv.Detections:
        results = self.model(image)
        return results.to_detections() # Convert your model's output to sv.Detections

2. Execute the model from the CLI

You can execute your model from the CLI. The parameters included into the __init__ method will be automatically added to the CLI. If the model class in inside the plugins folder:

modelzilla -i image.png -os plot MyModel --model_path model.pth

Otherwise, you need to specify the --plugins_folder argument:

modelzilla -i image.png -os plot --plugins_folder <path/to/your/plugin/folder> MyModel --model_path model.pth

Examples

Currently, we provide the following plugins:

• HFObjectDetection: A plugin that uses the HuggingFace Object Detection model.

How to execute it from the CLI

Install extra dependencies for the plugin:

pip install modelzilla[hf]

modelzilla -i http://images.cocodataset.org/val2017/000000039769.jpg -os plot HFObjectDetection --model_repo facebook/detr-resnet-50