Search
On April 29, the latest iteration of the Qwen series, Qwen3, a large language model, was officially unveiled and made open-source. This release encompassed a comprehensive range of dense and Mixture-of-Experts (MoE) models. Adhering to Haiguang Information's "Deep Intelligence" strategy, Haiguang DCU swiftly completed the seamless adaptation and fine-tuning of all eight Qwen3 models (235B, 32B, 30B, 14B, 8B, 4B, 1.7B, 0.6B). This achievement underscores second-level deployment with impeccable accuracy and zero compatibility issues. Qwen3 showcases remarkable inference performance and robustness on Haiguang DCU, solidifying the latter's versatility, ecological compatibility, and independently controllable technological prowess. Presently, Haiguang DCU stands as a pivotal infrastructure supporting AI large model training and inference.
-
C114 Communication Network -
Communication Home
