McGill-NLP/AfriqueQwen-14B

AfriqueQwen-14B: Enhanced Multilingual Model for African Languages

AfriqueQwen-14B is a 14 billion parameter causal language model from the AfriqueLLM suite, developed by McGill-NLP. It is built upon the Qwen3-14B-Base architecture and has undergone extensive continued pre-training (CPT) on approximately 26 billion tokens of multilingual data, including African monolingual data, code, and mathematics. This adaptation significantly improves its performance across 20 specific African languages, such as Swahili, Hausa, Yoruba, and Amharic, while preserving strong capabilities in high-resource languages like English and French.

Key Capabilities

• Multilingual Proficiency: Optimized for 20 African languages, demonstrating substantial performance gains over its base model on benchmarks like AfriMGSM, AfriMMLU, and FLORES (eng->xxx).
• Robust Base: Leverages the strong foundation of Qwen 3 models, which are noted for their performance preservation in high-resource languages and long-context tasks.
• Extensive Context: Supports a native context length of 32,768 tokens, enabling processing of longer texts.
• Diverse Training Data: Trained on a curated mix of African monolingual data, code (CornStack-Python), mathematics (FineMath-4+), and GPT-4.1 translated synthetic data.

Good For

• Applications requiring high-quality language understanding and generation in African languages.
• Tasks benefiting from long-context processing in a multilingual setting.
• Developers seeking a model with strong benchmark performance on African language evaluations.