Google recently announced Gemma, developed with the help of technologies used for Gemini. Developed by Google DeepMind and other internal teams, Gemma aims to provide support to artificial intelligence developers and researchers.
Google Gemma: What does it do?
Gemma comes in two versions: “Gemma 2B” and “Gemma 7B,” each with a lightweight structure. It’s worth noting that the AI language model can run on computers regardless of whether they are laptops or desktops, and it has IoT, mobile, and cloud support. Google collaborated with NVIDIA for this purpose.
According to reports, Gemma was trained on Google’s own datasets, thus posing no security risks. It’s also stated that feedback from humans during training was effective. Google encourages developers and companies interested in improving their own AI tools to use this AI model.
How good is Google Gemma?
Google also released data showing Gemma’s performance. The AI model achieved a 64.3% accuracy rate in a general Massive Multitask Language Understanding (MMLU) test covering 57 different topics. However, the results are somewhat less impressive in math tests. The model scored 46.4% accuracy in basic arithmetic operations taught in elementary school, and 24.3% accuracy in more complex math problems such as algebra and geometry.
{{user}} {{datetime}}
{{text}}