华为发布准万亿模型盘古Ultra MoE

Core Insights - Huawei has launched a new model called Pangu Ultra MoE with a parameter scale of 718 billion, marking it as a quasi-trillion MoE model [1] - The model was trained on the Ascend AI computing platform, and Huawei has released a technical report detailing the architecture and training methods of the Pangu Ultra MoE model [1] - Huawei has introduced key technologies that enable reinforcement learning (RL) post-training framework to enter the super-node cluster era on the Ascend CloudMatrix 384 super nodes [1] - The recently released Pangu Pro MoE model, with 72 billion parameters and 16 billion active parameters, ranks first domestically in the SuperCLUE leaderboard for models with parameters under 100 billion as of May 2025 [1]