Generative artificial intelligence (GenAI) models have long been seen as the domain of tech giants, mainly because of the immense resources required to train them. However, a recent breakthrough by PolyU researchers is set to challenge this perception. Their novel collaborative GenAI paradigm, known as Co-GenAI, significantly reduces the time and resources needed for training, while still achieving performance on par with leading models. This advancement has the potential to make cutting-edge AI research more accessible to institutions around the world.

Co-GenAI, developed by the PolyU Academy for Artificial Intelligence (PAAI), uses a decentralised approach to model training. This not only lowers resource barriers but also protects data privacy. The new framework enables ultra-low-resource training and decentralised model fusion, a method for combining different AI models, which has been validated through extensive real-world applications.

A key part of this breakthrough is the FP8 low-bit training solution developed by the team. PolyU is the first academic institution to open-source an end-to-end FP8 low-bit training solution that covers both continual pre-training and post-training stages. FP8 delivers over 20% faster training and reduces peak memory usage by over 10% compared to BF16 – another format used in AI training that requires more memory and time. In simple terms, researchers can train powerful AI models faster and with less computing power, making the process more efficient and affordable. The team also published the first theoretical validation of model fusion. The initial results of the research have been reported in academic publications.

Leading these groundbreaking advancements is Professor Yang Hongxia, Executive Director of PAAI, Associate Dean (Global Engagement) of the Faculty of Computer and Mathematical Sciences, and Professor of the Department of Computing. She stated, “Ultra-low-resource foundation model training, combined with efficient model fusion, enables academic researchers worldwide to advance GenAI research through collaborative innovation.”

Beyond theory, the team has applied their methods to real-world challenges across specific domains, such as developing top-performing medical AI models for cancer diagnosis and treatment. These models can adapt to medical devices for different scenarios, including personalised treatment and AI-based radiotherapy for oncology. Collaborations are underway with hospitals and research centres in Hong Kong and the Chinese Mainland, including the Queen Elizabeth Hospital, Huashan Hospital affiliated to Fudan University, Sun Yat-sen University Cancer Center, and Shandong Cancer Hospital to bring these innovations into clinical practice.

PAAI has also introduced an AI-powered academic writing assistant that supports a multimodal patent-search engine for research and manuscript drafting.

The research project is supported by the Research Grants Council’s Theme-based Research Scheme 2025/26, the Innovation and Technology Commission’s Research, Academic and Industry Sectors One-plus Scheme, and Cyberport’s Artificial Intelligence Subsidy Scheme. As Professor Christopher Chao, Senior Vice President (Research and Innovation) of PolyU, says, “These initiatives will not only solidify the leading position of PolyU in related fields, but also help position Hong Kong as a global hub for GenAI.”