China Claims World-First: Quantum Computer Fine-Tunes Billion-Parameter AI Model

World-first: Quantum computer tunes billion-parameter AI model

The experiment was conducted at the Anhui Quantum Computing Engineering Research Center, where this computer is operated. On this quantum chip, one set of data can run hundreds of tasks at the same time. That's a big deal for how fast AI models can learn. Furthermore, when this quantum chip was tested on a mental health chatbot dataset, the training improved by 15 percent. Mathematical accuracy also showed a spike, going from 68 percent to 82 percent. Even with the AI model size reduced by 76 percent, the performance still got better by 8.4 percent. This shows that quantum computers can help solve the problem of not having enough computing power to train huge AI models. Industry analysts said that this experiment has not only validated the feasibility of using quantum computing to achieve lightweighting of large language models (LLMs) but also laid a new path to address "computing power anxiety." Industry analysts have reported that this experiment opens the doors to achieving the light-weighting of large language models (LLMs) and the opportunity to address "computing power anxiety."

China 'Achieves' Quantum Breakthrough in AI Model Fine-Tuning

The experiment involved Origin Wukong, a third-generation superconducting quantum computer powered by a 72-qubit chip. Chinese scientists have successfully fine-tuned a billion-parameter AI model using the Origin Wukong superconducting quantum computer, claiming it to be the first of its kind worldwide. This feat was accomplished at the Anhui Quantum Computing Engineering Research Centre, as reported by Global Times. The experiment involved Origin Wukong, a third-generation superconducting quantum computer powered by a 72-qubit chip. It was a collaborative effort between Origin Quantum and other research institutions. The fine-tuning process reportedly involved adapting a large AI model for specialised applications by retraining it on domain-specific data. This was achieved by leveraging the quantum computer's ability to execute hundreds of parallel quantum tasks per batch. The results, as showcased by the research centre, indicated a significant improvement in model performance. A 76% reduction in parameters led to an 8.4% increase in training effectiveness, and the accuracy on a mathematical reasoning task rose from 68% to 82%. "It's like equipping a classical model with a quantum engine, allowing the two to work in synergy," explained Dou Menghan, vice president of Origin Quantum Computing Technology Co. The experiment highlights the potential of quantum computing to enhance AI model efficiency, particularly in reducing the computational resources required for large models. This breakthrough is a step forward in addressing 'computing power anxiety' associated with large AI models. Chen Zhaoyun, a researcher at the Institute of Artificial Intelligence, noted that this marks the first real-world application of quantum computing in large model tasks, proving the capability of existing hardware to support such operations. Origin Wukong has been operational since January 2024 and has completed over 350,000 quantum computing tasks across various industries, including fluid dynamics, finance, and biomedicine. Users from 139 countries have accessed the system remotely. In the broader context, this achievement demonstrates quantum computing's growing capabilities in supporting complex AI tasks. Integrating quantum and classical computing could lead to more efficient and powerful AI applications in fields like medical diagnosis and financial risk management.