Dynamic data sampler for cross-language transfer learning in large language models

TL;DR

This paper introduces ChatFlow, a cross-language transfer learning approach with a dynamic data sampler to efficiently train large Chinese language models using mixed-language corpora, improving convergence and performance.

Contribution

The paper presents a novel dynamic data sampler and a transfer learning framework that effectively trains Chinese LLMs by leveraging multilingual data and progressive training strategies.

Findings

Accelerates model convergence compared to baseline methods.

Achieves superior performance on Chinese and English benchmarks.

Outperforms other Chinese models post-trained on LLaMA-2-7B.

Abstract

Large Language Models (LLMs) have gained significant attention in the field of natural language processing (NLP) due to their wide range of applications. However, training LLMs for languages other than English poses significant challenges, due to the difficulty in acquiring large-scale corpus and the requisite computing resources. In this paper, we propose ChatFlow, a cross-language transfer-based LLM, to address these challenges and train large Chinese language models in a cost-effective manner. We employ a mix of Chinese, English, and parallel corpus to continuously train the LLaMA2 model, aiming to align cross-language representations and facilitate the knowledge transfer specifically to the Chinese language model. In addition, we use a dynamic data sampler to progressively transition the model from unsupervised pre-training to supervised fine-tuning. Experimental results demonstrate that our approach accelerates model convergence and achieves superior performance. We evaluate ChatFlow on popular Chinese and English benchmarks, the results indicate that it outperforms other Chinese models post-trained on LLaMA-2-7B.