Gemma 2

Top Builders

Explore the top contributors showcasing the highest number of app submissions within our community.

Gemma 2

Gemma 2 is the second-generation open large language model (LLM) from Google, built to advance performance, scalability, and responsible AI development. This model family supports diverse applications, from complex conversational AI to advanced content moderation systems. Based on Google’s leading transformer architecture, Gemma 2 builds on the Gemma foundation, incorporating enhanced safety, accessibility, and multi-platform deployment capabilities.

General
Relese date	February 2024
Author	Google DeepMind in collaboration with Google AI teams
Website	Google AI Gemma
Repository	Google AI Developer Resources
Type	Advanced open-source AI, large language model

Key Features:

Enhanced Parameter Options: Gemma 2 is available in configurations up to 27 billion parameters (2B, 9B, and 27B) for handling large-scale, complex language tasks.
ShieldGemma Safety Classifiers: ShieldGemma, a suite of safety-focused classifiers, helps detect and mitigate harmful content, addressing issues like hate speech, harassment, and explicit material.
Efficient Transformer Architecture: Supports processing up to 8192 tokens in a single pass, enhancing performance in long-form text processing and providing nuanced output across complex language tasks.
Flexible, Scalable Deployment: Optimized for both edge devices and cloud infrastructure, making it suitable for local, distributed, and high-demand deployment environments.
Integration with Leading AI Platforms: Works seamlessly with Google Cloud’s Vertex AI, Keras, JAX, and PyTorch, providing adaptable and efficient customization options for developers.

Applications:

Enterprise AI: Suitable for high-scale, complex applications such as automated data analysis, market predictions, and large-scale content creation.
Content Moderation Systems: The ShieldGemma classifiers filter harmful content, making it a good choice for moderation in social media, online communities, and customer service settings.
Multilingual Applications: With enhanced token processing, Gemma 2 is optimal for tasks requiring intricate language understanding and cross-language generation, ideal for global customer support and translation tools.
Conversational AI and Chatbots: Instruction-tuned models within Gemma 2 make it highly effective for advanced conversational systems, chatbots, and interactive voice assistants.

Get started building with Gemma 2:

You can start developing with Gemma 2 by accessing model weights from Google AI Studio and Kaggle. With integrated ShieldGemma safety tools, the model is prepped for responsible and large-scale deployments. The model’s compatibility with Google Cloud, Vertex AI, and popular AI frameworks makes it easy to customize for both edge and cloud solutions. Explore the full suite of resources on Google AI Gemma to unlock the potential of Gemma 2 and build robust, responsible AI applications.

Edit on GitHub

Google Gemma 2 AI technology Hackathon projects

Discover innovative solutions crafted with Google Gemma 2 AI technology, developed by our community members during our engaging hackathons.

CrosslaneAI

CrosslaneAI is an agentic migration platform that autonomously ports CUDA codebases to ROCm/HIP, so teams can run their GPU workloads on AMD hardware without rewriting a single kernel by hand. For over fifteen years, the entire GPU computing world has been written in CUDA. That accumulated codebase — not the silicon itself — is the real reason teams stay locked to a single vendor. AMD's MI300X is genuinely competitive today, offering 192GB of HBM3 and far better price-per-hour economics, yet most organizations can't make the switch. The migration is manual, slow, error-prone, and expensive enough that it's typically handed to specialist consultants charging weeks of engineering time. PortForge productizes that migration and collapses it from weeks into minutes.

Jaga — Cough-Based TB Triage

In 2024, around 2.4 million people who developed tuberculosis went uncounted—undiagnosed or unreported (WHO Global Tuberculosis Report 2025). Jaga, built by team Singkong, is an investigational phone-first triage prototype that helps close that gap. A community health worker records one guided cough and enters supported clinical inputs; Jaga returns a follow-up urgency estimate so programmes can prioritize who to send for confirmatory TB testing first. No X-ray machine, lab, or radiologist required. Two independent, never-fused AI pathways drive it. Gema handles cough plus clinical data: a Rust YAMNet gate confirms the audio is a cough, a local WavLM Large model embeds it entirely on-device, and XGBoost combines that with demographics. Prisma is a separate digital chest X-ray path (DenseNet121 + CLAHE) with its own estimate and a Grad-CAM heatmap. AMD is load-bearing: both models were trained on AMD Instinct MI300X via AMD Developer Cloud (ROCm PyTorch), and all model inference runs locally in Rust and Python. Fireworks serves only the Gemma guidance chat. Jaga does not diagnose or rule out TB. Every symptomatic person is directed to confirmatory testing—the estimate only prioritizes urgency.

Quantum Dot Live Twin

Quantum dot labs read device states via dispersive or RF measurements, operating much faster than the classification loops meant to track them. When a twin lags behind, drift can go unnoticed until valuable fridge time is lost recharacterising the device. We built a live digital twin designed to stay in sync with a quantum dot’s changing state under exactly that kind of pressure. It's a monitoring and synchronisation layer a lab can trust between tuning runs, not a system that wrestles the operator for gate control. The frames come from a real QArray-simulated device trajectory, powered by Rust for speed. In a 2000-frame run on an AMD MI300X (PyTorch/ROCm), the serial CPU baseline hits a worst-case lag of 0.155s. GPU batching without load-shedding actually performs worse under pressure, spiking to 3.304s as backlog builds toward the end. A rule-based triage agent solves this, holding worst-case lag to just 0.021s while cutting compute further, routing 235 full passes, 15 light passes, and 44 skips in real time based on queue depth, staleness, and drift signals, a decision actively used by the control loop. An LLM supervisor (Fireworks AI) sits above the triage agent, monitoring backlog trends in the background to retune thresholds without slowing the real-time path. The adaptive, LLM-tuned threshold architecture is running end-to-end. Next up is longer soak testing to capture threshold updates under sustained drift, which is a clear path to differentiation as the system scales to real hardware. The same stream is exposed as a true QCoDeS Instrument, so a lab can connect this twin directly to their instruments with no extra integration layer. That’s the difference between a demo and something a lab could plug in this week. Results arrive in an interactive control console built for the bench, not a slide deck. Future work includes moving from synchronous micro-batching to a fully asynchronous pipeline and training the classifier weights.

ReCodeX

ReCodeX is an AI-powered enterprise platform developed for the AMD Developer Hackathon: ACT II (Track 3 - Unicorn Track) designed to tackle the multi-billion dollar challenge of technical debt and legacy migration. Instead of a basic chatbot conversational layout, ReCodeX provides a complete engineering dashboard that automates the analysis, documentation, translation, and verification of aging infrastructure like legacy Java, COBOL, or unoptimized C++. The platform delivers comprehensive developer features, including: - Side-by-Side Modernization: Allows users to upload a legacy source file and view the original code side-by-side with refactored, modern, and fully test-covered code. - Automated Documentation: Provides deep semantic understanding of complex code logic to break down precisely what legacy components do. - Risk & Security Audits: Automatically generates a comprehensive Risk Report highlighting structural changes, security optimizations, and cloud-readiness metrics. The AMD Advantage & Tech Architecture: - To process heavy enterprise code demands reliably, ReCodeX relies entirely on high-performance cloud infrastructure backed by AMD hardware. - Core LLM orchestration is routed via the Fireworks AI API, utilizing high-performance AMD hardware endpoints running Google DeepMind's open-source Gemma models. - The compute layer is hosted on the AMD Developer Cloud, leveraging powerful cloud-based AMD GPUs to manage parallelized analytics, code scanning, and workflow pipeline tasks at sub-second latencies. - The entire system is packaged inside a Docker container for standardized deployment, with the frontend application built using Streamlit and React. Ultimately, ReCodeX targets tangible enterprise metrics for the Unicorn Track by reducing ongoing maintenance overhead, ensuring regulatory auditability through precise change logging, and accelerating cloud migration frictionlessly.

SimuFactory

Project Description: SimuFactory The Problem In the world of industrial automation, training technicians on live PLC (Programmable Logic Controller) systems is high-risk and high-cost. A single wrong command in a real factory can lead to catastrophic hardware failure, expensive production downtime, or severe safety hazards. However, traditional simulators are often static and fail to teach the "critical thinking" required for real-world troubleshooting. SimuFactory is a high-fidelity, AI-powered training ecosystem that implements a Cognitive Digital Twin of an industrial cell. Instead of a simple "correct/incorrect" feedback loop, SimuFactory utilizes a Socratic Coaching Engine. Powered by Llama-3, the AI acts as a senior engineer, guiding the trainee through complex faults using targeted questions and contextual hints, forcing the user to diagnose the root cause rather than simply following a manual. To ensure industrial-grade safety, we implemented a sophisticated Dual-LLM Architecture. While Llama-3 handles the pedagogical coaching, we integrated Gemma-2 as a dedicated Real-time Safety Auditor. This "Watchdog" model intercepts every terminal command, analyzing it against the current factory state to detect ISO 26262 safety violations. If a command is deemed dangerous, the auditor blocks the action and triggers a "Thermal Trip" or "E-Stop" simulation, teaching the trainee the real-world consequences of unsafe practices. SimuFactory is designed for Edge Deployment on AMD hardware. By leveraging the AMD ROCm™ stack and the massive memory bandwidth of the Instinct™ MI300X, we achieve ultra-low latency inference that allows the simulation and AI to react in real-time. Most importantly, our architecture ensures Data Sovereignty. In a real-world deployment, sensitive factory telemetry and PLC logic never leave the local network, removing the security risks associated with cloud-based AI and making SimuFactory a viable solution for high-security industrial environments

AI - Powered PCB Design

Revolutionize your electronics workflow with CircuitBlocks, the next-gen EDA software that uses advanced AI to automate PCB Design. Reduce your design-to-production time from weeks to hours with quick PCB generation. Built to accelerate hardware development by redefining the traditional EDA experience. By embedding machine learning directly into the layout and routing engine, the platform empowers engineers to shift their focus toward high-level system architecture while the AI manages the intricate, repetitive execution of Integrated Chip connections. Whether you are a solo hardware enthusiast or part of an enterprise engineering team, CircuitBlocks turns your technical intent into production-ready hardware with unparalleled efficiency.