Welcome to Sparsify

Sparsify improves model performance for deployment at scale using the latest model compression techniques. Use Sparsify to analyze, optimize, and integrate your neural network for production.

Sparsify is a deep learning model acceleration tool suite that simplifies the model optimization process to rapidly achieve the best combination of size, speed, and accuracy on any deep learning model. Sparsify is an easy way to optimize and benchmark models informed by industry research insights for ML practitioners, including ML engineers and operators, who need to deploy performant deep learning models at scale. Sparsify provides visual performance potential for your model, including a sliding scale between performance and loss.

About This Guide

This guide provides the following for you to use Sparsify:

  • Learning path and goals

  • Basic examples

  • Details of features

  • Advanced options

Feedback and Help

Sparsify is currently an alpha release and, as such, the software may not contain all of the features that are planned for the final version. Neural Magic, the software’s maintainer, welcomes all feedback and suggestions, from user experience to model optimization errors.

For support, sign up or log in to get help with your questions in our Sparsify channel : Discourse Forum and/or Slack .

Post software issues requiring support. Please include a light summary, the steps to reproduce, screenshots, files used, or onscreen messages. Be sure to share your environment details and framework versions.

GitHub Issues

Submit bugs and feature requests. Check existing issues to avoid duplication.


Next step…

Begin by installing Sparsify .