Jia Yangqing starts a new venture, building an AI legion that speeds up GLM-5.2 by 534%

Jia Yangqing, the creator of the deep learning framework Caffe and former VP at阿里巴巴, has embarked on a new entrepreneurial journey, forming an 'AI legion' of top talent. Their initial achievement is a dramatic acceleration of inference for the GLM-5.2 model, achieving a 534% speedup. This performance gain is realized through optimized inference techniques, and notably, the acceleration does not sacrifice model accuracy. The team's work focuses on enhancing the deployment efficiency of large language models, potentially reshaping competitive dynamics in the AI infrastructure space.