Google Gemma: Because Google doesn't want to give up on Gemini yet

Google has released Gemma 2b and 7b, a pair of open source AI models that allow developers to more freely use research conducted on its flagship Gemini phone. While Gemini is a large, closed AI model that competes directly with OpenAI's ChatGPT (and is almost as powerful), the lightweight Gemma is more likely to be suitable for smaller tasks like simple chatbots or summaries.

But what these models lack in complexity, they make up for in speed and cost of use. Despite their small size, Google claims that the Gemma models “outperform much larger models in key benchmarks” and are “capable of running directly on a developer's laptop or desktop computer.” It will be available via Kaggle, Hugging Face, Nvidia's NeMo, and Google's Vertex AI.

Releasing Gemma into the open source ecosystem is starkly different from how Gemini was released. While developers can build on Gemini, they do so either through application programming interfaces (APIs) or by working on Google's Vertex AI platform. Gemini is a closed model of artificial intelligence. By making Gemma open source, more people can experience Google's AI instead of turning to competitors that offer better access.

Both models will be available with a commercial license regardless of the size of the organization, the number of users, and the type of project. However, Google – like other companies – often prohibits the use of its models for specific tasks such as weapons development programs.

Gemma will also come with “responsible AI toolkits,” as it can be harder to put up guardrails in open models than in closed systems like Gemini. Trace Warkentin, director of product management at Google DeepMind, said the company had a “more extensive collaboration with Gemma because of the inherent risks involved in open models.”

The Responsible AI Toolkit will allow developers to create their own guidelines or list of banned words when deploying Gemma in their projects. It also includes a sample debugging tool that allows users to investigate Gemma's behavior and correct problems.

The models work best for English-related tasks at the moment, according to Warkentin. “We hope we can build with the community to meet market needs outside of English language assignments,” he told reporters.

Developers can use Gemma for free in Kaggle, and first-time Google Cloud users get $300 in credits for using templates. The company said researchers can apply for up to $500,000 in cloud credits.

While it's not clear how much demand there will be for smaller models like Gemma, other AI companies have released lighter versions of their main base models as well. Meta released Llama 2 7B, the smallest version of Llama 2, last year. The Gemini itself comes in several weights, including Gemini Nano, Gemini Pro, and Gemini Ultra, and Google recently announced the even faster Gemini 1.5 — again, for business users and developers right now.

By the way, Gemma means precious stone.