What AI says about NileChat

NileChat is a project by UBC NLP that develops a 3B-parameter language model adapted to Egyptian and Moroccan communities, aiming to reflect their language, cultural heritage, and values. It introduces a methodology to create synthetic and retrieval-based pre-training data tailored to a community’s language, culture, and values, demonstrated on Egyptian and Moroccan dialects. NileChat reportedly outperforms existing Arabic-aware LLMs of similar size and performs on par with larger models, and the project shares its methods, data, and models to promote inclusion of diverse communities in LLM development.