Gemma
An open-weight local LLM family for phones, PCs and servers, with Ollama and LM Studio support and a JAX library for fine-tuning.
Gemma is Google DeepMind's family of open-weight AI models for developers building applications on their own hardware. It includes compact models for mobile and IoT devices and larger models for reasoning on personal computers and servers.
- Reasoning and multimodal chat: Build applications that hold conversations, work with multimodal input, and support reasoning or AI agent workflows.
- Task-specific models: MedGemma handles medical text and images, TranslateGemma supports translation, and EmbeddingGemma produces on-device embeddings. FunctionGemma supports function calling on edge devices, while ShieldGemma classifies content for policy violations.
- Text generation: DiffusionGemma uses text diffusion with a focus on generation speed; T5Gemma provides encoder-decoder models for contextual understanding.
- Fine-tuning: The accompanying JAX library lets developers adapt Gemma models to their tasks, including through LoRA, and supports multi-turn, multimodal conversations.
Local integrations include Ollama, LM Studio, Gemma.cpp and Google AI Edge, with Android support also available. Gemma works with PyTorch and Keras. Offline applications are possible, including deployment on a local AI microserver; Google AI Studio and Google Cloud provide hosted options.
The JAX library is open source under Apache 2.0 and runs on CPU, GPU or TPU. That license applies to the library. Model size affects hardware requirements, with compact variants aimed at phones and larger models intended for personal computers and servers.
Logotyper
