Google launches Gemma, an open source lightweight AI model series

Google on Wednesday, February 21, released a new family of lightweight, open-source artificial intelligence (AI) models called Gemma. Two variants of Gemma, Gemma 2B and Gemma 7B, are 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 model. Interestingly, the Gemini 1.5 model was released last week. These smaller language models can be used to make task-specific AI tools, and the company allows responsible commercial use and distribution.

Google CEO Sundar Pichai announced the news at a press conference. postal on X (formerly Twitter). He said: “Gemma has demonstrated strong performance in benchmarks for language understanding and inference, is available globally starting today in two sizes (2B and 7B), supports a wide range of tools and systems, and is available on developer laptops , workstation or @GoogleCloud.”The company also owns Created Developer-focused AI model landing page, people can find quickstart links and code examples on their Kaggle model page, quickly deploy AI tools through Vertex AI, Google’s platform for building AI/ML tools for developers, or try Take the model and attach it to a separate domain using Collab (requires Keras 3.0).

Google highlighted some of the features of the Gemma AI model, saying both variants are pre-trained and tuned to instructions. It integrates with popular data repositories such as Hugging Face, MaxText, NVIDIA NeMo, and TensorRT-LLM. These language models can run on laptops, workstations, or Google Cloud with Vertex AI and Google Kubernetes Engine (GKE). The tech giant also released its new Responsible Generative AI Toolkit to help developers build safe and responsible AI tools.

According to reports shared by Google, Gemma outperforms Meta’s Llama-2 language model in multiple major benchmarks such as Large-Scale Multi-Task Language Understanding (MMLU), HumanEval, HellaSwag, and BIG-Bench Hard (BBH). Notably, Meta has already begun developing Llama-3, according to various reports.

Releasing small language models open source for developers and researchers has become a trend in the field of artificial intelligence. Stability, Meta, MosaicML, and even Google’s Flan-T5 model have been open source. On the one hand, it helps build an ecosystem where all developers and data scientists who don’t work with AI companies can try out the technology and create unique tools. On the other hand, this also benefits companies, as most offer their own deployment platforms that require a subscription fee. Additionally, developer adoption often highlights flaws in training data or algorithms that may have escaped detection before release, allowing businesses to improve their models.