Saved in:
Bibliographic Details
Main Authors: Singer, Philipp, Pfeiffer, Pascal, Babakhin, Yauhen, Jeblick, Maximilian, Dhankhar, Nischay, Fodor, Gabor, Ambati, Sri Satish
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2401.16818
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917639609122816
author Singer, Philipp
Pfeiffer, Pascal
Babakhin, Yauhen
Jeblick, Maximilian
Dhankhar, Nischay
Fodor, Gabor
Ambati, Sri Satish
author_facet Singer, Philipp
Pfeiffer, Pascal
Babakhin, Yauhen
Jeblick, Maximilian
Dhankhar, Nischay
Fodor, Gabor
Ambati, Sri Satish
contents We present H2O-Danube, a series of small 1.8B language models consisting of H2O-Danube-1.8B, trained on 1T tokens, and the incremental improved H2O-Danube2-1.8B trained on an additional 2T tokens. Our models exhibit highly competitive metrics across a multitude of benchmarks and, as of the time of this writing, H2O-Danube2-1.8B achieves the top ranking on Open LLM Leaderboard for all models below the 2B parameter range. The models follow core principles of LLama 2 and Mistral, and we leverage and refine various techniques for pre-training large language models. We additionally release chat models trained with supervised fine-tuning followed by direct preference optimization. We make all models openly available under Apache 2.0 license further democratizing LLMs to a wider audience economically.
format Preprint
id arxiv_https___arxiv_org_abs_2401_16818
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle H2O-Danube-1.8B Technical Report
Singer, Philipp
Pfeiffer, Pascal
Babakhin, Yauhen
Jeblick, Maximilian
Dhankhar, Nischay
Fodor, Gabor
Ambati, Sri Satish
Computation and Language
Machine Learning
We present H2O-Danube, a series of small 1.8B language models consisting of H2O-Danube-1.8B, trained on 1T tokens, and the incremental improved H2O-Danube2-1.8B trained on an additional 2T tokens. Our models exhibit highly competitive metrics across a multitude of benchmarks and, as of the time of this writing, H2O-Danube2-1.8B achieves the top ranking on Open LLM Leaderboard for all models below the 2B parameter range. The models follow core principles of LLama 2 and Mistral, and we leverage and refine various techniques for pre-training large language models. We additionally release chat models trained with supervised fine-tuning followed by direct preference optimization. We make all models openly available under Apache 2.0 license further democratizing LLMs to a wider audience economically.
title H2O-Danube-1.8B Technical Report
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2401.16818