Google He announced the renewal of his family of AI with open source, Gemma. New version, known as Gemma 3n, designed to work on mobile devices And you do not need to connect to the Internet. Google engineers optimized it to such an extent that Only a device with 2 GB of RAM is requiredField
According to the publication on the developers blog, Jamma 3n This is the evolution of Google open models. The new version based on the twins already has a multimodal capacity, which means that You can process images, audio and video. Gemma 3n also arrives with improvements in sections such as programming and reasoning, two characteristics that put it at the level of other large models.
As for specifications, Google engineers have done a lot The changes are focused on increasing performance on mobile devicesThe field is the most obvious adjustment is the change in the Matform, embedded in the transformer architecture, inspired by Matryoskas, which allows a larger model to contain smaller versions of itself. This technology allows you to perform each in accordance with the task, optimizing resources.
Gemma 3n also debuts a layer of inrocious products on a layer with which it can load the main weights in the VRM, and the rest in the processor. Unlike Gemma 3, which loaded all the parameters in the VRM, the new version administers them in such a way that Sharply reduces memory consumptionField
Gemma 3n can work on Android with a low level.
Google claims that Gemma 3n is designed from zero to function in mobile phones or devices with limited resourcesThe Family Field includes models: E2B, with 5000 million effective parameters and E4B, with 8,000 million parameters. The first requires at least 2 GB of RAM, and the second passes in terminals with 3 GB of RAM.
Other improvements include KS cache exchange, which improves response time in long contextual tasks, such as audio or video. According to tests, Gemma 3n twice as faster with the preparation of preliminary treatment against Gemma 3The Google field suggests that this is the first model with less than 10,000 million parameters exceeding 1300 points on the Lmarena test.
All of the above sounds very good on paper, but what does this mean for the user? Entry, the main advantage of the twins is that You do not need to connect to the Internet for workThe AI field can process the text, audio, images and video in real time and can transcrit a voice in the text. Models They are compatible with 140 languages for text and 35 for multimodal inputsamong which Spanish is included.
Users can Download gemma 3n via AI Edge GalleryThe Android application, which allows you to run the models directly on your mobile phone. From there you can try such tasks as analysis of images, text generation or voting transcription. You can also access Google Ai Studio or hug face examples through a browser.
Source: Hiper Textual
I’m Ben Stock, a highly experienced and passionate journalist with a career in the news industry spanning more than 10 years. I specialize in writing content for websites, including researching and interviewing sources to produce engaging articles. My current role is as an author at Gadget Onus, where I mainly cover the mobile section.
