Google DeepMind Unveils New 2 Billion Parameter Gemma 2 Model

www.analyticsdrift.com

Image source: Analytics Drift

On July 31st, 2024, Google DeepMind launched the novel 2 billion parameter model to the Gemma 2 family.

Image source: Google

Welcoming Gemma 2’s New Parameter Version

Google DeepMind posted on X that the 2 billion parameter Gemma 2 model can deliver top-tier performance despite its smaller size.

Image source: Google

Smaller Size, Bigger Impact

The new Gemma 2 parameter version can run efficiently on laptops, edge devices, and cloud platforms like Vertex AI and Google Kubernetes Engine.

Image source: Google

2B Model Operates on Diverse Hardware

According to DeepMind, the 2B Gemma 2 model is compact enough to operate on the free tier of the NVIDIA T4 deep learning accelerator.

Image source: NVIDIA

Runs Well on NVIDIA T4’s Free Tier

The 2B parameter version learns from large language models through knowledge distillation and offers exceptional results.

Image source: Canva

Improved 2B Model Performance

The team claims their 2B Gemma 2 version surpassed all GPT-3.5 models on the LMSYS Chatbot Arena Leaderboard.

Image source: Google

DeepMind’s 2B Model Tops GPT-3.5

DeepMind is also launching two new additions to the Gemma 2 series: ShieldGemma and Gemma Scope.

Image source: Huggingface

DeepMind Expands Gemma 2 Model Family