Cohere For AI, a research lab, has introduced Aya, a groundbreaking multilingual language model. This open-science initiative is one of the largest collaborative endeavours in machine learning, aiming to accelerate multilingual AI progress.
Aya is an instruction-finetuned open-access model that can follow instructions in a staggering 101 languages. Over 50% of these languages are considered lower-resourced, and Aya's performance is superior to leading models like mT0 and BLOOMZ in many tasks.
The model's performance is attributed to its ability to train on vast amounts of diverse data, which allows it to recognise patterns, words and structures easily. Aya's ability to understand multiple languages is also beneficial for effective communication and enhancing customer support.
Aya is preferred 77% of the time based on human evaluations, which is a significant achievement. The model has been trained with vast data sources and open-source datasets, which is an innovative approach to language models.
The introduction of Aya has brought us a step closer to the goal of a single universal model that can understand all languages, dialects and modalities and support speakers around the world. Aya is a powerful tool that broadens access to breakthroughs beyond English and other widely spoken languages.