More than 40 language and multimodal models completed pretraining on Huawei’s Ascend (昇腾) compute platform, which supports PyTorch.

The pretraining count comes from a Huawei executive’s statement. On September 19, Zhu Zhaosheng (朱照生), head of Huawei’s computing strategy and business development department, said more than 40 large language and multimodal models had completed pretraining using Ascend. The figure counts models, not accelerators, servers, or clusters.

Huawei also disclosed activity metrics for its software community. The Compute Architecture for Neural Networks (CANN) open-source community reportedly had more than 5,200 monthly active users. Zhu said non-Huawei developers outnumbered Huawei developers and that newly merged code averaged more than 30,000 lines per day.

Framework integration was presented alongside these ecosystem metrics. Zhu said Ascend had connected to mainstream open-source frameworks including PyTorch, in addition to its own open-source CANN software. He also claimed that Ascend was China’s only AI compute platform supporting pretraining and that its community had been the country’s most active since June; the supplied report provides no independent comparison supporting those rankings.

A separate announcement describes additional agent-software collaboration. At Huawei’s conference, 360 Group (三六零) reported a joint solution integrating its agent-factory infrastructure with Ascend AI for complex tasks, multi-agent collaboration, and long-chain reasoning. The announcement describes improved speed and stability but provides no numerical benchmark or customer-deployment scale.