Under-represented languages suffer from a lack of data, and as a result, there are few LLMs that support them. Extending an existing LLM to a new language is a practical option for startups, university labs, and organizations with limited budgets. This process involves several steps. In this paper, we describe how we adapted the Falcon3-7B model to Arabic, covering everything from data collection and training to evaluation. Falcon-Arabic was trained exclusively on native data to better capture the cultural and linguistic aspects of the language. Our evaluations show that Falcon-Arabic achieves state-of-the-art results on a range of Arabic benchmarks.

Arabic is one of the most widely spoken languages in the world, yet efforts to develop and evaluate Large Language Models (LLMs) for Arabic remain relatively limited. Most existing Arabic benchmarks focus on linguistic, cultural, or religious content, leaving a significant gap in areas like STEM and coding domains that are increasingly relevant for real-world LLM applications. To help bridge this gap, we present 3LM, a suite of three benchmarks designed specifically for Arabic. The first is a set of STEM-related question-answer pairs, naturally sourced from Arabic textbooks and educational worksheets. The second consists of synthetically generated STEM questions, created using the same sources. The third benchmark focuses on code generation, built through a careful translation of two widely used code benchmarks, incorporating a human-in-the-loop process with several rounds of review to ensure high-quality and faithful translations. We release all three benchmarks publicly to support the growth of Arabic LLM research in these essential but underrepresented areas.

Pre-trained language models are known to inherit a plethora of contextual biases from their training data. These biases have proven to be projected onto a variety of downstream applications, making their detection and mitigation imminent. Limited research has been conducted to quantify specific bias types, such as benevolent sexism, which may be subtly present within the inferred connotations of a sentence. To this extent, our work aims to: (1) provide a benchmark of sexism sentences; (2) adapt two bias metrics: mean probability score and mean normalized rank; (3) conduct a case study to quantify and analyze sexism in base and de-biased masked language models. We find that debiasing, even in its most effective form (Auto-Debias), solely nullifies the probability score of biasing tokens, while retaining them in high ranks. Auto-Debias illustrates a 90%-96% reduction in mean probability scores from base to debiased models, while only a 3%-16% reduction in mean normalized ranks. Similar to the application of non-parametric statistical tests for data that does not follow a normal distribution, operating on the ranks of predictions rather than their probability scores offers a more representative bias measure.