TensorX
返回文献探索

Paper · arXiv 2401.16818

H2O-Danube-1.8B Technical Report

Philipp Singer, Pascal Pfeiffer, Yauhen Babakhin, Maximilian Jeblick, Nischay Dhankhar, Gabor Fodor, Sri Satish Ambati

18 upvotesJanuary 30, 2024arXiv 预印本
AI 摘要

A 1.8B parameter language model trained on 1T tokens demonstrates competitive performance across benchmarks and is released publicly with supervised fine-tuning and preference optimization.

language modelpre-trainingsupervised fine-tuningdirect preference optimizationLLMs

Abstract

We present H2O-Danube-1.8B, a 1.8B language model trained on 1T tokens following the core principles of LLama 2 and Mistral. We leverage and refine various techniques for pre-training large language models. Although our model is trained on significantly fewer total tokens compared to reference models of similar size, it exhibits highly competitive metrics across a multitude of benchmarks. We additionally release a chat model trained with supervised fine-tuning followed by direct preference optimization. We make H2O-Danube-1.8B openly available under Apache 2.0 license further democratizing LLMs to a wider audience economically.

北京市昌平区探索星信息技术及软件开发工作室

京ICP备2026059466号
H2O-Danube-1.8B Technical Report | TensorX