Google has launched a new family of Artificial Intelligence models called Gemma and what he is thinking about to help developers and researchers create Artificial Intelligence models responsibly.
It is a model of open source AI that they can use to carry out their own developments. It is built with the same technology as Gemini, which also powers the chatbot and generational models of Gemini – formerly known as Bard – and the Artificial Intelligence tools of Google Workspace.
Gemma presents two different models: Gemma 2B and Gemma 7B. Both have been trained and understand thousands of instructions and variables that can be executed on a laptop or non-desktop computer with a CPU or GPU.
According to Google, Gemma is much more efficient and more rigorous than other AI models. For example, it is trained to filter personal or sensitive information, and uses people's feedback to prevent the development of potentially harmful tools. Of course, Gemma at the moment only offers support for text actions, not like Gemini, which is multimodal and does provide support for image or video.
At the moment, Gemma can be accessed for free on Kaggle and Colab. It can also be accessed through Hugging Face, MaxText and Nvidia NeMo.
With the presentation of this development, just a week after having launched the Gemini update, which is called Gemini 1.5 and which improves the multimodal AI tool, Google continues in a fierce race with OpenAI to lead the development of Artificial Intelligence in all fields.
This last company also presented Sora, its AI for creating video, last week, although at the moment it has not released open source versions for its models. It is expected that it will imitate Google and do so soon to continue in the race.