FuxiTranyu: A Multilingual Large Language Model Trained with Balanced Data

TL;DR

FuxiTranyu is a balanced, multilingual large language model with 8 billion parameters, trained on diverse data covering 43 natural and 16 programming languages, achieving competitive performance and consistent representations across languages.

Contribution

The paper introduces FuxiTranyu, a new open-source multilingual LLM trained on balanced data, with instruction tuning and alignment, advancing multilingual capabilities and interpretability.

Findings

FuxiTranyu outperforms existing multilingual LLMs on various benchmarks.

The model demonstrates consistent multilingual representations across languages.

Open-source release facilitates further research in multilingual NLP.

Abstract

Large language models (LLMs) have demonstrated prowess in a wide range of tasks. However, many LLMs exhibit significant performance discrepancies between high- and low-resource languages. To mitigate this challenge, we present FuxiTranyu, an open-source multilingual LLM, which is designed to satisfy the need of the research community for balanced and high-performing multilingual capabilities. The base model, FuxiTranyu-8B, features 8 billion parameters and is trained from scratch on meticulously balanced multilingual data that contains 600 billion tokens covering 43 natural languages and 16 programming languages. We also develop two instruction-tuned models: FuxiTranyu-8B-SFT which is fine-tuned on a diverse multilingual instruction dataset, and FuxiTranyu-8B-DPO which is further refined with DPO on a preference dataset for enhanced alignment ability. Extensive experiments on a wide range of multilingual benchmarks demonstrate the competitive performance of FuxiTranyu against existing multilingual LLMs, e.g., BLOOM-7B, PolyLM-13B, and Mistral-7B-Instruct. Both neuron and representation interpretability analyses reveal that FuxiTranyu achieves consistent multilingual representations across languages. To promote further research into multilingual LLMs, we release both the base and instruction-tuned FuxiTranyu models together with 58 pre-training checkpoints at HuggingFace (see https://huggingface.co/TJUNLP/FuxiTranyu-8B) and Github (see https://github.com/tjunlp-lab/FuxiTranyu).