Averroes-Q-Instruct: A Bilingual Arabic-English 7B Instruction-Tuned Model for the Hayula Ecosystem

We present Averroes-Q-Instruct (ابن رشد), a bilingual Arabic-English 7-billion parameter instruction-tuned language model built on Qwen2.5-7B-Instruct. The model was fine-tuned on 55K human-written instructions spanning both Arabic and English, using Low-Rank Adaptation (LoRA) via the MLX framework on a single Apple M2 Ultra workstation with 192GB unified memory. Averroes-Q-Instruct achieves a final training loss of 1.434, demonstrating effective bilingual instruction following without catastrophic forgetting of the base model's capabilities. The model serves as the foundational base for the Saif (سيف) multi-expert cybersecurity AI suite, providing bilingual Arabic-English competence that enables specialized cybersecurity models to operate across both languages. Named after the Andalusian polymath Ibn Rushd (Averroes), the model embodies the principle of bridging intellectual traditions. We release Averroes-Q-Instruct under the Apache 2.0 license at BinSaqban/Averroes-Q-Instruct and hayulalab/Averroes-Q-Instruct on HuggingFace.