Google DeepMind today announced the release of the 2 billion (2B) parameter version of Gemma 2, the second generation of its Gemma AI models.
First launched in February this year, Gemma is a family of lightweight, open text-to-text models for developers and researchers – and is based on the technology that underlies Google Gemini.
DeepMind released Gemma 2 in June in two different sizes: 9 billion (9 billion) and 27 billion (27) parameters.
The new 2B model learns by distilling larger models and delivers above-average results, says DeepMind. The company also claims that it outperforms all GPT-3.5 models on the LMSYS Chatbot Arena leaderboard.
TNW Conference 2025 – Back to NDSM on June 19-20, 2025 – Save the date!
Having concluded our incredible 2024 edition, we are delighted to announce our return to Amsterdam NDSM in 2025. Register now!
Gemma 2 2B can run on a wide range of hardware, from laptops to edge devices and cloud deployments using Vertex AI and Google Kubernetes Engine (GKE). Additionally, it is small enough to run on the free tier of the NVIDIA T4 deep learning accelerator.
ShieldGemma and Gemma Scope
DeepMind is also introducing two further additions to the model family: ShieldGemma and Gemma Scope.
ShieldGemma is a set of security classifiers designed to detect and moderate malicious content in the inputs and outputs of AI models. It comes in different sizes and targets hate speech, harassment, sexually explicit content, and dangerous content.
GemmaScope focuses on transparency. The tool includes a collection of sparse autoencoders (SAEs), which are specialized neural networks that decode the complex inner workings of Gemma 2 models and provide an easier-to-understand format for information processing and decision making.
There are over 400 freely available SAEs covering all layers of Gemma 2 2B and 9B. The goal is to enable researchers to develop more transparent and reliable AI systems.
Starting today, developers and researchers can download Gemma 2 2B from Kaggle, Hugging Face, and Vertex AI Model Garden, or try it out in Google AI Studio. ShieldGemma and Gemma Scope are available here and here.
Comments are closed.