Google has launched Gemma, a new AI model targeting researchers who want to run AI models on devices rather than on cloud services.
According to Google, the new Gemma model was developed by its British subsidiary DeepMind in cooperation with other teams within the company.
The Gemma model shares some "technical and infrastructure components" with the broader Gemini model, and Google offers two models: the Gemma 2B and the Gemma 7B.
Researchers have access to pre-trained Gemma models that are configured to receive instructions and can run locally on desktops and laptops or via the cloud.
Google has expanded its new model to support NVIDIA graphics cards as well as Google Cloud TPUs, which are Google processors that run AI tasks in Google Cloud.
Google says Gemma is technically smaller than other large language models (LLMs), but says it "significantly outperforms it in key metrics," including Meta's Llama 2.
Google is now making Gemma available to researchers for free through Kaggle and Colab, and is also offering researchers free credits to use the model in the cloud via Google Cloud.
Earlier this month, the company launched the Gemini model with a free app of the same name and a $20 paid subscription for premium features. Later, the company launched the Gemini 1.5 Pro model with more features than the first version.