H2O-Danube3 Technical Report

TL;DR

H2O-Danube3 introduces a series of compact, high-performing language models trained on web data, optimized for mobile deployment, and openly available to democratize access to advanced LLM technology.

Contribution

The paper presents a new series of small, efficient language models trained on large web datasets, optimized for mobile use, and openly released under an open-source license.

Findings

Models achieve competitive benchmarks across multiple tasks.

Designed for efficient inference on smartphones.

Openly available to promote democratization of LLMs.

Abstract

We present H2O-Danube3, a series of small language models consisting of H2O-Danube3-4B, trained on 6T tokens and H2O-Danube3-500M, trained on 4T tokens. Our models are pre-trained on high quality Web data consisting of primarily English tokens in three stages with different data mixes before final supervised tuning for chat version. The models exhibit highly competitive metrics across a multitude of academic, chat, and fine-tuning benchmarks. Thanks to its compact architecture, H2O-Danube3 can be efficiently run on a modern smartphone, enabling local inference and rapid processing capabilities even on mobile devices. We make all models openly available under Apache 2.0 license further democratizing LLMs to a wider audience economically.