Zhipu, a Chinese AI startup, has used Huawei Ascend chips for its all-new AI multimodal model. The news is emerging on the internet at times when Nvidia is planning to enter the Chinese AI market and tempt consumers with its H200 processor.
Also known as Knowledge Atlas Technology JSC Ltd., Zhipu has recently launched a new multimodal AI model, packed with Huawei Ascend chips over US techs.
The company said its latest creation is China’s first AI model to be fully trained on native chips rather than relying on foreign technologies or any other US products.
Speaking of the model, it is an open-source image generation AI model named GLM-Image. It uses the Ascend Atlas 800T A2 server + MindSpore framework, packed with various Huawei Kunpeng and Ascend AI chips.
Zhipu said:
“This proves the feasibility of training high-performance multimodal generative models on a domestically developed full-stack computing platform.”
Huawei Ascend chips are now becoming a choice for many startups in the Chinese market. And one such is Zhipu, which has launched the first ‘state of the art’ multimodal platform of China using the efficient Ascend processors.
Good news is that no other firms except for Zhipu and Huawei in China have trained their AI models successfully on domestically-produced chips to date.
Just like Huawei, Zhipu has also struggling with the US sanctions since last year. Instead of giving up, the company made efforts, and now, its shares are seeing a sharp rise of 80% in China’s AI industry.
If Chinese clients continue to cut ties with US technologies and adopt self-made AI chips for their products, it would be a great help to the country amid the US-China tech tensions.
(Image Credits: Huawei)
|| source ||
The post China’s Zhipu AI model chooses Huawei Ascend over US chips appeared first on Huawei Central.
