google-gemma/awesome-gemma
A curated map of everything around Google's Gemma open model family, in one list
awesome-gemma is a link-collection repository that organizes resources around Gemma, Google DeepMind's family of lightweight open models. It groups links by category: core models and variants, ways to run Gemma locally or in the cloud, fine-tuning tools, tutorials, demo apps, hackathon projects, and even research papers. Rather than code, the repository itself is a curated index pointing to official docs, Hugging Face collections, inference engines, and community projects.
What it does
- Lists core Gemma 4 models plus specialized variants like MedGemma (medical), PaliGemma 2 (vision-language), TranslateGemma (55 languages), VaultGemma (trained with differential privacy), and Gemma Scope 2 (interpretability tooling).
- Covers ways to run Gemma both locally (via tools like llama.cpp, Ollama, vLLM, JAX, Docker) and through hosted cloud services (Google Cloud, OpenRouter, Cerebras, NVIDIA, AMD, and others).
- Points to fine-tuning resources including the official Gemma Cookbook, Unsloth's training guide, and Apple Silicon/MLX-specific tuning tools.
- Showcases community demos such as browser-based agents using WebGPU, a car damage inspection app, and hackathon-winning projects from the 'Gemma 4 Good Challenge' addressing healthcare and disaster response.
- Notes that Gemma has even been run in orbit, with Starcloud-1 executing it on an H100 GPU in space and NASA using it for satellite image analysis.
Why it matters
For anyone starting to work with Gemma models, this list saves time by consolidating scattered official docs, tools, and community projects into one navigable reference. It also reveals the breadth of Gemma's ecosystem, from tiny on-device deployments to specialized domains like medicine and space, helping readers gauge what's mature and where opportunities exist.
Terms in this repo
- Gemma · Google DeepMind이 공개한 경량 오픈 모델 계열
- QAT (Quantization-Aware Training) · 모델을 저정밀도로 변환해도 성능 저하가 적도록 훈련 단계에서 양자화를 반영하는 기법
- differential privacy · 개별 데이터가 결과에 미치는 영향을 제한해 프라이버시를 보호하는 훈련 방식
- sparse autoencoder · 모델 내부 표현을 해석 가능한 단위로 분해하는 해석성 연구 도구
- speculative decoding · 작은 모델로 먼저 예측하고 큰 모델이 검증해 생성 속도를 높이는 기법
Repository description (English)
😎 Awesome list about Gemma, Google DeepMind's family of lightweights, state-of-the-art open models.
Open on GitHubTrending repos
- cathrynlavery/diagram-designA skill that makes AI coding tools draw magazine-quality diagrams instead of generic rounded boxes
- public-apis/public-apisA giant crowd-curated directory of free APIs for developers
- semantica-agi/semanticaAn open-source graph infrastructure that lets AI agents show their work, not just their answers
- cactus-compute/needleA 14MB AI model small enough to run tool-calling on a phone or watch, without internet
- unslothai/unslothA desktop app that lets you run and train AI models on your own computer, no coding required
- macro-inc/macroAn all-in-one workspace where email, chat, docs, tasks, and CRM are cross-linked and share one AI memory
- harry0703/MoneyPrinterTurboAn open-source tool that turns a single topic or keyword into a finished short video, complete with script, footage, subtitles, and music
- basecamp/omarchyA ready-made, opinionated Linux setup built by DHH
Latest from METAL LAB
- Open-source Ornith-1.5 claims scores on par with Claude Opus
- Modular, Now Owned by Qualcomm, Fully Open-Sources Mojo Language
- Anthropic Launches Free "Academy" Site for Learning Claude
- Liquid AI's 300M Draft Model Speeds Up Decoding by Up to 3.18x
- ChatGPT sites now handled by Codex instead of Git and CI, adds team invites