Artificial Intelligence thread

tokenanalyst

Brigadier
Registered Member

Yushu Technology plans to file for an IPO in Q4 this year, with quadruped robots accounting for 65% of its revenue.​


Hangzhou Unitree Technology Co., Ltd. (hereinafter referred to as "Unitree") officially announced on the social media platform "X" that it is actively advancing preparations for its initial public offering (IPO). The company expects to submit IPO filing documents to the stock exchange between October and December 2025, at which time it will officially disclose relevant operating data.

1757000278167.png

Yushu Technology also revealed in a statement that by 2024, quadruped robots will account for 65% of the company's revenue, humanoid robots will account for 30%, and the remaining 5% will come from the sale of related components and parts. Approximately 80% of quadruped robots will be used in research, education, and consumer applications, while the remaining 20% will be used in industrial applications such as inspection and firefighting. Humanoid robots will be used entirely in research, education, and consumer applications.

Since its founding, Yushu Technology has been committed to the application of high-performance general-purpose robots in various civilian industries. This commitment is clearly stated and addressed on the company's official website, product manuals, partnership agreements, and various other documents. We urge all parties to exercise caution and not mistake other companies' robotics products or third-party modified devices for Yushu Technology products. We hope that Yushu Technology's robots will bring safer and more enjoyable lives to people around the world.

Please, Log in or Register to view URLs content!
 

tokenanalyst

Brigadier
Registered Member

"Supernode" ecosystem accelerates integration: SenseTime and Huawei Ascend achieve key adaptations to jointly create a domestic AI computing power foundation​


SenseTime's large-scale SenseCore processor and Ascend 384 supernode recently completed full integration, achieving expected results in both functional and performance verification. This collaboration marks a key advancement in system-level collaboration and engineering implementation of domestic AI computing power, propelling domestic high-performance computing architecture from "usability" to "performance," providing a more stable and efficient computing foundation for large-scale model training and inference.

Superpods, a key form of AI computing infrastructure, integrate multiple NPUs/GPUs into a unified computing unit through high-speed interconnection, aiming to overcome bottlenecks in computing power coordination and communication efficiency in large-scale model training. Huawei's Ascend 384 Superpod, with its "fully peer-to-peer" architecture, achieves system-level resource pooling across servers and cabinets, interconnecting CPUs, NPUs, DPUs, storage, and other components at high speed, delivering computing power density and bandwidth comparable to supercomputers.

1757000715605.png1757000816189.png

SenseTime and Huawei Ascend collaborate to achieve multiple innovations in super-node adaptation


At the same time as Huawei Ascend was launched, this new solution architecture also put forward higher requirements for software stack upgrades and platform scheduling optimization, allowing it to "run fast and stable."

As an AI cloud-native platform, SenseCore is committed to providing users with agile, flexible, and reliable full-stack AI infrastructure services, promoting the efficient implementation and large-scale application of large-model technology with extreme cost-effectiveness.

Leveraging the unique features of SenseCore, SenseTime's large-scale equipment, and Ascend's 384 supernode, the two teams collaborated to develop a number of industry-leading innovations in scheduling optimization, system stability, and fault recovery.
1757000755028.png
  • Scheduling optimization: In terms of scheduling capabilities, in addition to supporting basic capabilities such as single-machine and multi-machine scheduling within a POD, multi-machine scheduling across PODs, and affinity scheduling, the SenseCore platform cooperates with model parallel strategies to achieve automatic division of logical super nodes, allowing large communication strategies such as EP/TP to fully utilize the Lingqu network and improve model training efficiency.​
  • Cross-Pod Training Stability: In addition, the SenseCore team submitted multiple MRs to fix the master/worker task rank disorder problem in multi-Pod scenarios, fundamentally solving the problem of probabilistic failure of cross-Pod training tasks.​
  • Multi-dimensional fault detection and recovery: The Ascend 384 supernode provides multi-dimensional fault detection, encompassing everything from server hardware, high-speed interconnects, and the RoCE network to tasks, processes, and software and hardware. This combined detection capability enables a multi-level recovery mechanism for jobs, pods, and processes, comprehensively improving the reliability and fault tolerance of the Ascend 384 supernode in training scenarios.​
The successful integration of SenseCore, a large-scale device, and Ascend 384 supernodes has made multi-tenant, large-scale, and elastic AI cloud services possible. Going forward, the two companies will explore additional application scenarios, including large-model inference acceleration, intelligent agent application deployment, and large-model training and inference optimization for vertical industries, further accelerating the adoption of SenseCore-based Ascend 384 supernodes across various industries.

Please, Log in or Register to view URLs content!
Please, Log in or Register to view URLs content!
 
Top