Google releases Gemma, a family of lightweight open-source AI models for developers

Google released a new lightweight, open-source family of artificial intelligence (AI) models called Gemma on Wednesday, February 21. Two variants of Gemma, Gemma 2B and Gemma 7B, were made available to developers and researchers. The tech giant said it used the same technology and research for Gemma that it used to create the Gemini AI models. Interestingly, the Gemini 1.5 model was introduced last week. These smaller language models can be used to create artificial intelligence tools for specific tasks, and the company allows responsible commercial use and distribution.

The announcement was made by Google CEO Sundar Pichai in a post of X (formerly known as Twitter). He said: “Demonstrating strong performance in language understanding and reasoning benchmarks, Gemma is available worldwide from today in two sizes (2B and 7B), supports a wide range of tools and systems and runs on a developer laptop, workstation or @GoogleCloud.” The company has too created a developer-oriented landing page for the AI ​​model where people can find quickstart links and code examples on the Kaggle Models page, quickly implement AI tools through Vertex AI (Google’s platform for developers to build AI/ML tools), or to play around with the model and attach it to a separate domain using Collab (will require Keras 3.0).

Highlighting some of the features of the Gemma AI models, Google said that both variants are pre-trained and set up with instructions. It is integrated with popular data stores such as Hugging Face, MaxText, NVIDIA NeMo and TensorRT-LLM. Language models can run on laptops, workstations or Google Clouds via Vertex AI and Google Kubernetes Engine (GKE). The tech giant also released a new Responsible Generative AI Toolkit to help developers build safe and responsible AI tools.

According to reports shared by Google, Gemma outperformed Meta’s Llama-2 language model in multiple benchmark tests such as Multitasking Language Understanding (MMLU), HumanEval, HellaSwag, and BIG-Bench Hard (BBH). It should be noted that Meta has already started work on Llama-3, according to various reports.

Releasing smaller language models open source for developers and researchers is something that has become a trend in the AI ​​space. Stability, Meta, MosaicML and even Google with its Flan-T5 models already exist in open source. On the one hand, it helps build an ecosystem, as all the developers and data scientists who don’t work with the AI ​​firms can try their hand at the technology and create unique tools. On the other hand, it also benefits the company, since most often the companies themselves offer implementation platforms that come with a subscription fee. Additionally, developer adoption often highlights flaws in the training data or algorithm that may have escaped detection prior to release, allowing enterprises to improve their models.