Hala Technical Report: Building Arabic-Centric Instruction & Translation Models at Scale
This work addresses the need for specialized Arabic NLP models, which is a domain-specific problem for researchers and practitioners in Arabic language processing, though it is incremental as it builds on existing translation and fine-tuning techniques.
The authors tackled the problem of creating high-quality Arabic instruction and translation models by developing Hala, a family of Arabic-centric models trained using a translate-and-tune pipeline that compresses a teacher model to FP8 for efficiency and generates a million-scale Arabic instruction corpus. The result was state-of-the-art performance on Arabic-centric benchmarks in both nano (≤2B) and small (7-9B) parameter categories, with specific gains such as ~2× higher throughput without quality loss.
We present Hala, a family of Arabic-centric instruction and translation models built with our translate-and-tune pipeline. We first compress a strong AR$\leftrightarrow$EN teacher to FP8 (yielding $\sim$2$\times$ higher throughput with no quality loss) and use it to create high-fidelity bilingual supervision. A lightweight language model LFM2-1.2B is then fine-tuned on this data and used to translate high-quality English instruction sets into Arabic, producing a million-scale corpus tailored to instruction following. We train Hala models at 350M, 700M, 1.2B, and 9B parameters, and apply slerp merging to balance Arabic specialization with base-model strengths. On Arabic-centric benchmarks, Hala achieves state-of-the-art results within both the "nano" ($\leq$2B) and "small" (7-9B) categories, outperforming their bases. We release models, data, evaluation, and recipes to accelerate research in Arabic NLP.