Dynamic Data Sampler for Cross-Language Transfer Learning in Large Language Models

• Main Authors: Li, Yudong, Feng, Yuhao, Zhou, Wen, Zhao, Zhe, Shen, Linlin, Hou, Cheng, Hou, Xianxu
• Format: Conference Proceeding
• Language: English
• Subjects: Computational modeling; cross-language; knowledge transfer; Large language model; Natural language processing; Protocols; Reproducibility of results; Signal processing; Training; Transfer learning
• ISSN: 2379-190X
• DOI: 10.1109/ICASSP48485.2024.10446640

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.