The 2024 Conference on Empirical Methods in Natural Language Processing

In recent years, Large Language Models (LLMs) have made significant strides towards Artificial General Intelligence. However, training these models from scratch requires substantial computational resources and vast amounts of text data. In this paper, we explores an alternative approach to constructing a LLM for a new language by continually pre-training (CPT) from existing pre-trained LLMs, instead of using randomly initialized parameters. Based on parallel experiments on 40 model sizes ranging from 40M to 5B parameters, we find that 1) CPT converges faster and saves significant resources in a scalable manner. 2) CPT adheres to an extended scaling law derived from with a joint data-parameter scaling term. 3) The compute-optimal data-parameter allocation for CPT markedly differs based on our estimated scaling factors. 4) The effectiveness of transfer scale is influenced by training duration and linguistic properties, while robust to data replaying, a method that effectively mitigates catastrophic forgetting in CPT. We hope our findings provide deeper insights into the transferability of LLMs at scale for the research community.

Thank you for providing detailed and insightful feedback on our paper. We are very pleased to see the positive evaluation from the reviewers and to see the recognition of our contribution in the field of Cross-Lingual Continual Pre-Training (CPT).

In response to your suggested revisions, we reply as follows:

Tokenization and its Impact on CPT: In the revised version of our paper, we will discuss the tokenizer used in our experiments and consider potential impacts of vocabulary extension on CPT performance.

Color Scheme of Figure 1: Thank you for your feedback on the color scheme of Figure 1. We will reassess and improve the color scheme to enhance clarity.

We will carefully consider your valuable suggestions and implement these improvements in the revision process of our paper. Thank you again for your positive and constructive feedback.