The IA Gemma 3 model is energy efficient and could turn on relatively inexpensive equipment. Previously already, Openai had introduced a smaller version of its AI models.
The new AI model has 270 million parameters. Gemma 3 270m, as his name is, is the little brother of other mini-models Gemma 3 from Google. Beforehand, the company had indeed launched versions characterized by 1 to 27 billion parameters. These parameters indicate the power of a model, but also the amount of energy and calculation power it requires. Gemma 3 270m would be able to turn locally or in the cloud and, thanks to its low imprint, it could work entirely on a smartphone or in a browser. ‘Even on a toaster’, specifies the developer Omar Sanseviero on the social network X.
The idea underlying this model is that companies could use it to create several specialized models, each fulfilling a specific task. Google itself evokes in particular the analysis of feelings and compliance controls. The Gemma 3 models are ‘open’, according to the terminology of technological giant, which is not the same with open source. You can download the models for free and adjust certain weights yourself to configure them. The developers are however invited to respect the instructions for use.
While the main developers of AI often seem to compete in order to offer the most parameters and possible calculation power on an AI model, the accent is also placed on smaller versions, capable of operating outside cloud. These have the advantage of being better suited to work sensitive to confidentiality and of being often more efficient. This is how previously already, Openai released smaller models, comprising 120 and 20 billion parameters respectively, capable of filming on a laptop (powerful). Llama 3.2 of Meta is available in versions at 1 and 3 billion parameters. Gemma 3 270m is still much smaller in this list. By way of comparison, a ‘large’ model as chatgpt4 has some 1.8 billiard of parameters.
OPENAI introduces an AI model capable of running on a laptop
