Atlas-Chat: Adapting Large Language Models for Low-Resource Moroccan Arabic Dialect

TL;DR

Atlas-Chat is introduced, the first-ever collection of LLMs specifically developed for dialectal Arabic, and it is believed this work offers comprehensive design methodologies of instruction-tuning for low-resource languages, which are often neglected in favor of data-rich languages by contemporary LLMs.

Abstract

We introduce Atlas-Chat, the first-ever collection of LLMs specifically developed for dialectal Arabic. Focusing on Moroccan Arabic, also known as Darija, we construct our instruction dataset by consolidating existing Darija language resources, creating novel datasets both manually and synthetically, and translating English instructions with stringent quality control. Atlas-Chat-2B, 9B, and 27B models, fine-tuned on the dataset, exhibit superior ability in following Darija instructions and performing standard NLP tasks. Notably, our models outperform both state-of-the-art and Arabic-specialized LLMs like LLaMa, Jais, and AceGPT, e.g., our 9B model gains a 13% performance boost over a larger 13B model on DarijaMMLU, in our newly introduced evaluation suite for Darija covering both discriminative and generative tasks. Furthermore, we perform an experimental analysis of various fine-tuning strategies and base model choices to determine optimal configurations. All our resources are publicly accessible, and we believe our work offers comprehensive design methodologies of instruction-tuning for low-resource languages, which are often neglected in favor of data-rich languages by contemporary LLMs.

Figures (9)

• Figure 1: Average model scores over all benchmarks.
• Figure 2: A Sample from TÜLU-V2-mix.
• Figure 3: The prompt given to Claude 3.5 Sonnet for translation.
• Figure 4: Hard coded instruction-answer pairs.
• Figure 5: The prompt given to Claude 3.5 Sonnet for translation and cultural adaptation of the AlpacaEval instructions.