Alibaba Cloud said July 27 that its Lingjun Zhenwu M890 supernode instances were adapted to run Kimi K3, a large language model.

The company said tests showed that joint optimization across chips, the inference platform, and the model cut first-token latency by about 35%.

First-token latency measures how quickly a large language model starts showing output, so the improvement should make long-context interactions smoother.

Alibaba Cloud added that its Qianwen AI Platform and Bailian would provide application programming interfaces, or APIs, for Kimi K3.

It also said both sides would deepen cooperation on domestic computing power.

🔗 Source: Security Times

Recent Alibaba Cloud developments