Google launches AI model “Gemma”

Feb 22, 2024 | NEWS, AI, GOOGLE

Gemma: AI Model

According to Reuters, on Wednesday, Google unveiled a new open-source AI model named Gemma, which external developers can adopt and customize, mirroring a strategy previously employed by Meta and other companies.

Google launches AI model "Gemma

Alphabet stated that individuals and companies can develop artificial intelligence programs using the new AI model from the open family of models,” explained Kellogg. Key technical data, such as model weights, is made available to the public.

This initiative is expected to draw software engineers to leverage Google technology and promote the utilization of its cloud division. The company noted that this AI’s models are primarily optimized for Google Cloud, with first-time users of the service receiving a $300 account credit.

While Google has not yet fully open-sourced Gemma, implying that the company might retain a role in determining terms of use and ownership, it does offer developers the ability to establish their own guidelines or a prohibited word list when incorporating Gemma into their projects. Additionally, this model includes an error correction tool, enabling users to scrutinize Gemma’s behavior and rectify issues.

According to The Virgin, these models currently excel in tasks related to the English language, with Google aiming to collaborate with developers and the AI community to address the requirements of other language markets.

Unlike Gemma, Google has not opened up its larger, distinct models known as Gemini, and stated that Gemma’s models comprise 2 to 7 billion parameters, representing the various values the algorithm considers when generating output.

Meta’s artificial intelligence models vary in size from 7 to 70 billion parameters, while Google has not disclosed the dimensions of its largest Gemini models.

OpenAI’s GPT-3 model, announced in 2020, comprises 175 billion parameters.

Infidia revealed yesterday that it collaborated with Google to optimize the performance of Gemma models on its chips. Additionally, Infidia announced plans to soon adapt its ChatPot program, designed for running artificial intelligence models on Windows computers, to work seamlessly with Gemma.

 

You can also see:

Sora: Introducing a Cutting-Edge AI Model for Text-to-Video Conversion