Gemma Model: A New LLM Model that is Lightweight and “Open”

Gemma Model: Lightweight and Claimed to be “Leading”

Just a week after the latest iteration of their Gemini model. Google has just announced Gemma, a new family of lightweight and “open” LLM (Large Language Model) models. The Gemma 2B and 7B models are inspired by Gemini and are available for commercial and research use.

Although Google has not provided detailed comparisons with similar models from Meta and Mistral, they claim both are “cutting-edge” in technology. The Gemma model uses a dense decoder-only architecture, similar to previous Gemini and PaLM models. Benchmarks will be available on the Hugging Face leaderboard soon.

Accessibility and Use of the Gemma Model

Developers can use Gemma through ready-to-use Colab and Kaggle notebooks, as well as integration with Hugging Face, MaxText, and Nvidia’s NeMo. Once trained and tuned, this model can run on various platforms.

Although Google refers to Gemma as an “open” model, it should be noted that this model is not open source. At a press conference, Google’s Jeanine Banks emphasized the company’s commitment to open source but also explained that they use the term “open” carefully.

According to her, the term “open model” is now commonly used, but it often refers to models with open weights, where developers and researchers can customize and improve the model. However, usage requirements, such as redistribution and ownership of developed variants, may vary depending on the specific model’s terms.

Google chose the term “open” because developers can use the model for inference and refinement freely. The Google team argues that the size of this model is suitable for various use cases.

Advantages and Potential of the Gemma Model

Tris Warkentin, Director of Product Management at Google DeepMind, said, “The quality of production has significantly improved in the past year. Things that were previously only possible with very large models can now be done with small cutting-edge models.

This opens up exciting new ways to develop AI applications, including running inference and tuning on developers’ local desktop or laptop with their RTX GPU, or on a single host on GCP with Cloud TPU.”

Impact of the Launch

The launch of the Gemma model by Google could have several impacts, including:

  1. Accelerating LLM Adoption: The lightweight and “open” Gemma model may encourage more developers and companies to adopt LLM in various applications. This could accelerate the development and use of AI technology in various fields.
  2. Increasing Competition in the LLM Field: Google’s entry into the “open” LLM market could increase competition with other companies such as Meta and Mistral. This competition could drive innovation and the development of better and cheaper LLM models.
  3. Strengthening AI Security and Responsibility: The responsible generative AI toolkit launched alongside Gemma could help developers build safer and more responsible AI applications. This could increase public trust in AI technology.
  4. Opening New Opportunities for AI: It could open up new opportunities for AI applications in various fields, such as:
    • Education: To develop more personalized and adaptive learning platforms.
    • Healthcare: Assisting doctors in diagnosing diseases and developing more personalized treatment plans.
    • Finance: To detect fraud and develop more effective investment strategies.
    • Manufacturing: Assisting in improving manufacturing efficiency and productivity.
  5. Strengthening Google’s Position in AI: The launch of the Gemma model could strengthen Google’s position as a leader in AI technology. This could help Google attract top talent and increase its market share in the AI field.

Also Read: Adobe’s New AI Assistant for Reading PDF