Google’s New Gemma 3 Open-Source AI Models Can Run on a Single GPU

Google released the Artificial Intelligence (AI) model of the GEMMA 3 family on Wednesday. The successor of the Jemma 2 series, introduced in August 2024, comes with the new open-source model text and visual argument capabilities. Mountain view-based tech veteran stated that these models provide support for more than 35 languages ​​and may be fine to provide additional assistance to 140 languages. In particular, the tech giant claimed that these models have been adapted to run on single GPU or Google’s Tensor Processing Unit (TPU).

Google Gemma 3 series AI model releases model

One in blog postTech giant expanded the new Small Language Model (SLM), which highlighted that they were designed using the company’s major Gemini 2.0 model. The Gemma series is known to open-sound and offer on-device performance. Google has revealed that so far, the GEMMA model has been downloaded more than 100 million times and has been used to create more than 60,000 variants.

Tech veteran claimed that Jemma Meta’s Lama-405B, Dipsek-V3, O 3-Mune AI model of OpenEE improves LMARANA’s leadersboard on LMARANA’s leadersboard. These models are available in four sizes – 1B, 4B, 12B and 27B criteria. They can be run on single GPU or TPU, the company claimed.

The GEMMA 3 chain offers advanced text and visual argument abilities, and can analyze images, text and short videos. The AI ​​models offer a reference window of 1,28,000 tokens. Model function calling also provides calling support, which will allow developers to manufacture agent capabilities in apps and software that they make.

Google said the AI ​​model was developed using careful risk evaluation. The company claimed that it implements internal security policies through fine-tuning and benchmark evaluation. The model was also tested using a more competent model. The model displayed low risk levels, claiming the company.

In addition to the GEMMA 3 series, the company also launched Shildgemma 2, a 4B parameter image security checker, which ensures that the AI ​​models do not generate materials with dangerous, sexually explicit or violent materials. Developers have also been given the option to adapt shieldgemma to increase safety parameters. AI model’s Gemma 3 family can be downloaded through Google’s hug of Google Entry Or on Kagal.