Google DeepMind announced the release of its Gemma 2, a 2 billion parameter version of the AI model family. Gemma is a lightweight text-to-text open model designed for developers and researchers. The new model learns from larger models through distillation and produces better results than GPT-3.5 models on the LMSYS Chatbot Arena leaderboard.
Gemma 2 can run on various hardware, including laptops, edge devices, cloud deployments with Vertex AI and Google Kubernetes Engine (GKE), and even on the free tier of the NVIDIA T4 deep learning accelerator.
The company is also introducing ShieldGemma, a series of safety classifiers designed to detect and moderate harmful content in AI model inputs and outputs. Gemma Scope focuses on transparency, providing an easier-to-understand format of how the Gemma 2 models process information and make decisions.
There are over 400 freely available sparse autoencoders covering all layers of Gemma 2 2B and 9B, allowing researchers to create 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 on their respective pages.
Source: https://thenextweb.com/news/google-deepmind-2b-parameter-gemma-2-model