Gemma: Small, Powerful, and Redefining Open Source AI Ethics

Author(s): Jelle Van Bost

Originally published on Towards AI.

Google’s new model offers impressive performance alongside a unique licensing approach designed to promote innovation and responsible use.
Image generated by the author on https://leonardo.ai/

Google’s new open-source model, Gemma, is shaking up the landscape of Large Language Models (LLMs). Despite its size (with 7B and 2B parameter variants), Gemma’s performance is remarkable, rivaling the performance of larger open-source models. This innovation raises important questions: How can such a small and powerful model be used responsibly? And how can users be held accountable for their actions? Google chose to release Gemma with a unique, protective license, which could signal a new way forward within the open-source community.

The model performs better than other open-source models

Image from “Gemma: Open Models Based on Gemini Research and Technology”

Gemma performs exceptionally well on Question Answering and Reasoning and even outperforms other models on Math/Science and Coding benchmarks! Google proved that the 7B model improves upon other existing 7B parameter models. But what about the 2B model?

Image from “Gemma: Open Models Based on Gemini Research and Technology”

The 2B parameter model performs worse on most tasks, still outperforming LLaMA-2 7B on certain tasks (SIQA 0-shot, CAQ 7-shot, HumanEval pass@1, MBPP 3-shot, GSM8K maj@1, MATH 4-shot and BBH). The impressive thing about Gemma is that it can even run on your laptop, according to Google.

So far, Gemma… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI