Chinese Researchers Achieve Breakthrough in AI Model Post-Training with Huawei Chips

Chinese chipmakers have found success in supporting AI inference but struggled with training. Researchers, including Huawei, successfully conducted full-parameter post-training on a 1.6 trillion-parameter model using over 1,000 Huawei chips. This enhances China's AI self-reliance. The project, jointly conducted with the Shenzhen Loop Area Institute, Harbin Institute of Technology (Shenzhen), and Shenzhen Research Institute of Big Data, used a computing cluster to update the model's entire architecture. Previously, domestic computing power was used primarily for inference. This breakthrough adds complexity to AI capabilities, multiplying computational demands. It will enhance China's AI industry chain self-reliance, according to a Shenzhen government social media post.