Extending LLMs to New Languages: A Case Study of Llama and Persian Adaptation

TL;DR

This paper investigates extending Llama, a large language model, to support Persian through parameter-efficient fine-tuning, multi-stage training, and bilingual data alignment, demonstrating improved Persian task performance with minimal impact on English tasks.

Contribution

It introduces a multi-stage approach for adding Persian to Llama, combining monolingual pretraining, bilingual alignment, and instruction-tuning, highlighting the effectiveness of bilingual data in low-resource language adaptation.

Findings

Incorporating Persian improves classification accuracy for Persian tasks.

Bilingual data alignment enhances low-resource language performance.

Cross-lingual transfer has limited benefits for Persian with minimal training data.

Abstract

Large language models (LLMs) have made great progress in classification and text generation tasks. However, they are mainly trained on English data and often struggle with low-resource languages. In this study, we explore adding a new language, i.e., Persian, to Llama (a model with a limited understanding of Persian) using parameter-efficient fine-tuning. We employ a multi-stage approach involving pretraining on monolingual Persian data, aligning representations through bilingual pretraining and instruction datasets, and instruction-tuning with task-specific datasets. We evaluate the model's performance at each stage on generation and classification tasks. Our findings suggest that incorporating the Persian language, through bilingual data alignment, can enhance classification accuracy for Persian tasks, with no adverse impact and sometimes even improvements on English tasks. Additionally, the results highlight the model's initial strength as a critical factor when working with limited training data, with cross-lingual alignment offering minimal benefits for the low-resource language. Knowledge transfer from English to Persian has a marginal effect, primarily benefiting simple classification tasks.