Google has been promoting its various Gemini AI models for the past few weeks. Today, it revealed a new and smaller model for a specific user type. The new model is being called Gemma and its audience is those researchers who want to use a local model rather than access AI via the cloud.
In a blog post, Google stated that Gemma was developed by its UK-based DeepMind along with other teams at the company. Gemma does share some "technical and infrastructure components" with its bigger brother Gemini. There are two specific models available: Gemma 2B and Gemma 7B. Researchers can access "pre-trained and instruction-tuned Gemma models" that can run on their local desktop or laptop, as well as in the cloud. It will be optimized to run on Nvidia GPUs as well as Google Cloud TPUs (Tensor Processing Units).
Google claims that while Gemma 2B and 7B models are technically smaller compared to other large language models (LLMs), it says they surpass "significantly larger models on key benchmarks" including Meta"s Llama 2.
Google is making Gemma available now for researchers for free in its Kaggle researcher platform, and it is also available as a free tier Colab notebooks. If you have yet to use Google Cloud, you can access Gemma with $300 free credits, and researchers can purchase up to $500,000 in credits to use it via Google Cloud.
Earlier this month, Google announced that its Bard chatbot would be renamed as Gemini. It also launched a mobile Gemini app for Android, and offered access to the AI model via the Google search app for iOS devices. The company also lanched Gemini Advanced at the same time, which is its most advanced model for the price of $19.99 a month.
Later in February, the company announced Gemini 1.5 Pro, which is supposed to be more powerful than both 1.0 Pro and 1.0 Ultra. There"s no word on a release date for that.