In recent weeks, most of the talk has been about Gemini at Google. A new AI model from Google that is supposed to become the engine of all AI projects. Today Google Deepmind suddenly presents an open model called Gemma. Open in this case means that the model can be used anywhere. Hugging Face is one of the companies that will start working directly with the model.
Gemma is based on the same techniques and technology as Gemini. According to Google, Gemma’s performance is better than Meta’s LLAMA 2 models. This involves the knowledge of the model and its ability to solve problems. For this, they ran an awful lot of different benchmarks.
An open AI Model?
With an open model, Google wants to help developers build their own responsible AI solutions. At the same time as Gemma, Google is also introducing the Responsible Generative AI Toolkit. This tookit allows developers to debug their AI but also learn best practices based on Google’s experience. Gemma is available in the Vertex AI Model Garden, as well as from Kaggle Models and Hugging Face. As an open model, Gemma can be used anywhere.
Gemma comes in two variants, a 2B and a 7B. Both models are light enough to run on a laptop. However, the 7B model is a lot more powerful. However, it is also more potent than the 13b model of LLAMA 2, which allows us to conclude that the size of the AI models are starting to get smaller. The latter is a good thing because the models become more efficient, and less hardware is needed to use them.
Open Source community leading the way?
Despite all the intelligent developers and data geniuses at companies like Google, OpenAI and Microsoft, the open-source community still manages to impress. These companies put a lot of money into developing these kinds of models. They lack the funds in the open source community, but with knowledge and ingenuity, they sometimes manage to get more done. For example, experts in the open-source community regularly achieve better results with small, well-trained models. That remains extraordinary when you consider how many people and resources those tech companies have invested.
This ensures that big tech giants also remain motivated to keep sharing their knowledge and models in the AI field. After all, the AI race has only just begun, and a tremendous amount of innovation is expected in the coming years. So, it could just be that the open-source community provides the next breakthrough based on a model like Gemma. For Google and others, it will be a good learning opportunity to get even more out of the models.
23 hours ago
