Google Releases New Family of AI Language Models Called Gemma for Desktop Computers
Google has recently announced the launch of Gemma, a new family of AI language models that are characterized by being free and open-weights. These models are designed to be operational locally on desktop or laptop computers, unlike the more powerful but restricted Gemini models.
The Gemma family consists of two versions: Gemma 2B, with 2 billion parameters, and Gemma 7B, with 7 billion parameters, both of which are available in pre-trained and instruction-tuned formats. Parameters in AI refer to the values that dictate the behavior of an AI model, with a portion of these parameters known as weights being stored in files.
Developed by Google DeepMind and other Google AI teams, the Gemma project builds upon the insights gained during the creation of their Gemini models, which serve as Google’s leading commercial large language models (LLM) efforts. These models power features like the Gemini AI assistant.
The name “Gemma” is derived from Latin and means “precious stone”.
The launch of Gemma represents a significant move by Google in the open large language model space, following the success of OpenAI’s ChatGPT. It also aligns with Google’s history of contributing to open AI research with technologies such as the Transformer architecture, TensorFlow, BERT, T5, and JAX.
While Gemma is not classified as “open source” due to specific licensing connotations, it is seen by experts as a crucial step towards improving transparency and privacy in chatbots. Google’s move seems to be a strategic effort to compete against Meta and its release of open-weights models, aiming to attract more developers to Google’s Vertex AI cloud platform.
According to Google, the Gemma 7B model outperforms Meta’s Llama 2 models in various benchmarks, particularly in math, Python code generation, general knowledge, and commonsense reasoning tasks. Currently, Gemma is accessible through platforms such as Kaggle and Hugging Face.
In addition to the Gemma models, Google has also introduced a “Responsible Generative AI Toolkit” that aims to provide developers with resources to create safe and responsible AI applications.
Analyst comment
Positive news.
As an analyst, the launch of Google’s Gemma family of AI language models for desktop computers is a significant move that improves transparency and privacy in chatbots. Google aims to compete against Meta and attract more developers to their cloud platform. Gemma’s superior performance in various benchmarks positions it as a strong contender in the open large language model space. This development, combined with the introduction of a Responsible Generative AI Toolkit, demonstrates Google’s commitment to providing safe and responsible AI applications. Expect increased adoption of Gemma models and traction for Google’s cloud platform.
