Shulou(Shulou.com)11/24 Report--

CTOnews.com January 7 news, the Great Wall's self-driving company Momo Zhihang announced at the scene of the Seventh HAOMO AI DAY, the intelligence center MANA OASIS (Snow Lake Oasis) officially established.

According to reports, MANA OASIS is currently the largest intelligence center in China's self-driving industry, jointly built by Mo Zhixing and volcano engine, with 6.7 billion floating-point operations per second, storage bandwidth of 2T per second and communication bandwidth of 800G per second.

Gu Weihao, CEO of Mimo Zhixing, said that after two years of research and development, they have established a full set of Data Engine for large-scale training in data management capabilities, achieving a 10-fold increase in 100 P data filtering speed and a random read and write delay of 10 billion small files less than 500 microseconds.

In terms of computing power optimization, they also deployed a Lego high-performance operator library, ByteCCL communication optimization capabilities, and a large model training framework.

In terms of training efficiency, based on Sparse MoE, hundreds of billions of parameter large model training can be easily completed through cross-machine sharing, and the training cost of millions of Clips (minimum video annotation unit) only needs 100 carat weeks, and the training cost is reduced by 100 times.

In addition, Hao Mo also launched five major MANA models, including the visual self-supervision model (the first automatic annotation of 4D Clip in China) Reduce labor costs by 98%), 3D reconstruction of large models (solving data distribution problems with lower costs), multi-modal mutual supervision large models (can complete the identification of general obstacles), dynamic environment large models (reduce dependence on high-precision maps), self-supervised cognitive models of human driving (make driving strategies more anthropomorphic, safe and smooth).

It is said that the millimeter city NOH is expected to land in 100 cities in the first half of 2024, which is the first urban navigation-assisted driving in China that can land on a large scale, keeping a leading position in the direction of re-perception, technical route and application of large models, and data construction of closed-loop users. CTOnews.com has learned that officials also say they will fully enter the self-driving era of HPilot in 2025.

Zhang Kai, chairman of Mo Zhihang, also summed up their achievements in 2022. It is reported that the third generation of HPilot products are loaded with nearly 20 models, such as Wei brand, tank, Euler, Great Wall Gun, etc., helping driving users to drive more than 25 million kilometers. Millimeter City NOH is the first mass-produced urban navigation-assisted driving product in China. At present, the software seal has reached the delivery state. The terminal logistics automatic distribution vehicle has initially completed the commercial closed loop, delivering more than 1000 units, the small magic camel distribution order has exceeded 130,000, and the commercialization process is speeding up in an all-round way; the MANA learning time is more than 420000 hours, and the virtual driving age is equivalent to 55,000 years of human drivers.

In addition, it is not only the first company in China to develop its own AEB algorithm at home and abroad, but also the first self-driving company in China to go to sea in Europe and Australia.

He believes that the competition for the second half of smart driving will enter the acceleration period in 2023, and the commercial applications of high-end intelligent driving products will be launched on a large scale.

At the industry level, self-driving in 2023 will fully enter the 3.0 era; high-level intelligent driving will become the standard of mid-range models; a new generation of chips and sensors will enter the market, and the autopilot ecology will show more diversity.

At the technical level, the ability of large models in data synthesis and knowledge extraction will improve the iterative speed of autopilot, and the competition of autopilot system will change from functional competition to commuting efficiency competition. the supercomputing center will become an entry configuration for autopilot enterprises.

At the product level, urban navigation-assisted driving has entered the re-perception stage, and the curtain of large-scale mass production and delivery has been lifted; the user experience of intelligent driving will change from tasting freshness to user dependence; and the overall cost of the terminal logistics automatic distribution vehicle will be reduced to less than 100000 yuan. will further improve production relations.

In the field of talent, Zhang Kai judged that the fierce competition for talent in the AI autopilot field will continue until 2023.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.