This repository hosts quantized TensorFlow Lite (.tflite) versions of Google's Gemma models, specifically for use in Android sample applications, hackathons, and educational projects.

The goal is to provide a frictionless, anonymous download path for on-device integration without requiring app users to log in or developers to embed secret API keys.

The original Gemma models are created by Google and are officially hosted on platforms like Kaggle and the official Google Hugging Face account.

These official repositories are often "gated," meaning they require a user to be logged in and to have explicitly accepted the license terms before a download can start. This is a necessary compliance step for Google to ensure users agree to their Responsible AI terms.

However, this authentication requirement prevents direct, anonymous downloads from within a mobile application and would create a poor user experience (e.g., asking an app user to log in to a developer service).

This repository acts as a public, non-gated mirror. The models are redistributed here under the permissions granted by their Apache 2.0 License. This allows mobile demo apps to fetch the model file via a direct URL without any user-side authentication, while still respecting the original license terms.

The models here are intended for developers who need to quickly integrate a Gemma model into an Android application for demonstration or prototyping purposes.

They can be downloaded directly within an app using Android's DownloadManager or a networking library like OkHttp, providing a seamless "first-run" experience for the end-user.

All models in this repository are the property of Google and are governed by the Apache 2.0 License and the Gemma Terms of Use.

Official Gemma Terms of Use: https://ai.google.dev/gemma/terms

⚠️ IMPORTANT: If you use these models in your application, you must include proper attribution to Google and provide a copy of or link to the license in your app, as required by the license terms.