AQL-Lite framework for efficient Arabic LLM adaptation

AQL-Lite uses a quality-first curation pipeline and RS-LoRA adapters to adapt multilingual LLMs to Modern Standard Arabic. Using a 1B-token budget, the method achieves a 3.51 point improvement in Arabic benchmarks over standard continual pretraining on raw web data.