Xpeng Motors launches research and development of a 72 billion parameter self-driving model

16 Apr Xpeng Motors launches research and development of a 72 billion parameter self-driving model

London, April 16, 2025, (Oilandgaspress) –––– The automotive industry has entered a new wave of large-scale model development. On April 14, Xiaopeng Motors held an AI technology sharing meeting in Hong Kong, revealing for the first time that it is developing a large-scale autonomous driving model with 72 billion parameters, namely the “Xiaopeng World Base Model”. In the future, Xiaopeng will deploy the base model to the vehicle side by distilling the small model in the cloud, and equip the “AI car” with a new brain. This model will also empower Xiaopeng’s AI robots, flying cars, etc.

Li Liyun, head of autonomous driving at Xpeng Motors, said that the Xpeng base model is a multimodal large model with a large language model as the backbone network and trained with massive amounts of high-quality driving data. It has visual understanding, chain reasoning and action generation capabilities. Through powerful reinforcement learning training, the base model continues to evolve and will gradually develop autonomous driving technology that is comparable to or even beyond that of humans.

Li Liyun revealed that in order to develop the base model, Xiaopeng Motors began to lay out AI infrastructure as early as last year . It has now built the first Wanka intelligent computing cluster in the domestic automotive industry to support pre-training, post-training, model distillation, and vehicle-side model training of the base model. Xiaopeng Motors calls this cloud-to-end production process the “cloud model factory”. At present, Xiaopeng’s “cloud model factory” has a computing power of 10 EFLOPS, and the cluster operation efficiency remains above 90% all year round. The full-link iteration cycle from cloud to end can reach an average of once every 5 days.

Relying on powerful AI infrastructure and data infrastructure, Xiaopeng has launched a new base model research and development paradigm. From cloud model pre-training to vehicle-side model deployment, the iteration cycle of the entire “cloud model factory” is an average of once every 5 days.

Information Source: Read More

Oil and gas press covers, Energy Monitor, Climate, Gas,Renewable, Oil and Gas, Wind, Biomass, Sustainability, Oil Price, LPG, Solar, Marine, Aviation, Fuel, Hydrogen, Electric ,EV, Gas,