model-archiver

What it is and what it does

Model Archiver is a command-line tool that packages trained neural network models into `.mar` archive files suitable for inference with MXNet-Model-Server. It accepts a model file, a handler script, and metadata, then bundles them into a standardized archive format that MMS can consume and serve. The tool is lightweight, depending only on future and enum-compat for compatibility.

The package is part of the broader MXNet-Model-Server (MMS) ecosystem but can be installed standalone. It was designed to streamline the workflow of taking a trained model from development into a serving-ready format. However, the project is now archived with no active maintenance since mid-2024, and the latest release is from 2019, making it a legacy tool that may not align with current deep-learning deployment practices or MXNet versions.

Use it for:

• Package a trained PyTorch or MXNet model into a `.mar` file for deployment with MXNet-Model-Server.
• Automate model archiving in a CI/CD pipeline to prepare models for inference serving.
• Bundle a model with a custom inference handler to define preprocessing and postprocessing logic.
• Create reproducible model archives for sharing trained models across teams or environments.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Model Archiver creates `.mar` archive files from trained neural network models for use with MXNet-Model-Server inference.

No. Although install friction is low and the license is permissive, the project is abandoned (archived repository, no releases since 2019, last commit 2024-05-20). MXNet-Model-Server itself is no longer actively maintained, and this tool is unlikely to work reliably with modern Python versions or current model-serving frameworks. Consider modern alternatives like TorchServe, TensorFlow Serving, or other actively maintained model-packaging tools.

Install

model-archiver on PyPI

pip install model-archiver

uv add model-archiver

poetry add model-archiver

Installing model-archiver

Before you install

Install friction is low; the package is a pure-Python wheel with only two lightweight runtime dependencies (future and enum-compat). However, the repository is archived and the latest release dates to 2019-05-16, with no commits since 2024-05-20—this is an abandoned project with no active maintenance.

License in practice

Licensed under Apache License Version 2.0 (permissive), so you may use, modify, and distribute it freely in commercial and private projects, provided you include a copy of the license and state any changes made.

Quickstart

pip install model-archiver

model-archiver archive -m model_name -s model.pt -e handler.py -o model.mar

ONNX support requires manual installation of protobuf compiler, onnx, and mxnet; the tool itself does not declare these as dependencies.

Verify before relying

• Whether the tool remains compatible with current MXNet versions or modern Python releases.
• Whether `.mar` files created by this version remain compatible with current MXNet-Model-Server deployments.
• Specific Python version compatibility (requires_python is unspecified in the fact sheet).

Package facts

Evidence: model_archiver-1.0.3-py2.py3-none-any.whl

Keywords: MXNet, Model, Archive, Archiver, MMS, Server, Serving, Deep, Learning, Inference, AI

Tags

More Artificial Intelligence packages