sleepstory-gemma-models

Gemma models (int4, MediaPipe .task format)

This repo mirrors Gemma checkpoints packaged as MediaPipe LLM Inference API .task containers, for use by Sleep Story — a personal, offline Android app that runs the model fully on-device. The app lets a user choose which model to download; this repo hosts the variants it offers.

Files

Provenance

• Base models: google/gemma-3n-E2B-it (https://huggingface.co/google/gemma-3n-E2B-it) and google/gemma-3-1b-it (https://huggingface.co/google/gemma-3-1b-it)
• Format: converted to the MediaPipe .task container and int4 quantized, following Google's published LiteRT/MediaPipe conversion path (see litert-community on Hugging Face for equivalent pre-converted builds).
• Modification notice: these files are a modified form of the original Gemma weights — repackaged into the .task container format and quantized to int4. They are not bit-identical to the upstream checkpoints.

Why this mirror exists

The official .task builds are hosted behind Hugging Face's gated-access flow, which requires an authenticated request per download. This mirror is an unauthenticated, plain-HTTPS copy so a resumable range-request download can run from the app without embedding a user token. Distribution here follows Section 3.1 of the Gemma Terms of Use (see LICENSE and NOTICE in this repo).

License

Use of these models is governed by the Gemma Terms of Use (LICENSE in this repo, mirrored from https://ai.google.dev/gemma/terms), including the Gemma Prohibited Use Policy (https://ai.google.dev/gemma/prohibited_use_policy), which is incorporated by reference. By downloading or using these files you agree to be bound by both.