Gemma 2: Improving Open Language Models at a Practical Size

TL;DR

Gemma 2 introduces lightweight open language models with innovative architectural modifications and knowledge distillation, achieving state-of-the-art performance at smaller sizes and providing accessible models to the community.

Contribution

It presents Gemma 2, a family of efficient open language models with architectural enhancements and knowledge distillation, outperforming larger models.

Findings

Gemma 2 models achieve top performance for their size.

Models are competitive with 2-3 times larger models.

All models are publicly released for community use.

Abstract

In this work, we introduce Gemma 2, a new addition to the Gemma family of lightweight, state-of-the-art open models, ranging in scale from 2 billion to 27 billion parameters. In this new version, we apply several known technical modifications to the Transformer architecture, such as interleaving local-global attentions (Beltagy et al., 2020a) and group-query attention (Ainslie et al., 2023). We also train the 2B and 9B models with knowledge distillation (Hinton et al., 2015) instead of next token prediction. The resulting models deliver the best performance for their size, and even offer competitive alternatives to models that are 2-3 times bigger. We release all our models to the community.