行业动态
x.com ·
@AI at Meta (@AIatMeta)
Meta 披露 Prometheus 千兆瓦级超算训练可靠性数据
Meta 表示其 Muse 系列基础模型（含 Muse Wonder）在俄亥俄州 New Albany 的千兆瓦级超算集群 Prometheus 上训练。该集群拥有数十万块 GPU，Meta 称通过面向大规模韧性的工程设计，实现预训练 goodput 85%+、强化学习 90%+，Muse Wonder 训练期间平均约 47K GPU 小时才出现一次中断。
AI推荐理由 Meta 官方账号披露 Prometheus 千兆瓦级超算集群的训练可靠性指标：预训练 goodput 85%+、强化学习 90%+、Muse Wonder 训练期间平均约 47K GPU 小时才发生一次中断。这是少见的超大规模训练基础设施一手数据，对算力与训练工程有参考价值；但仅为推文摘要，缺少指标定义、统计口径和可验证细节，且 Muse 模型本身信息有限。
78
浏览 228,178
点赞 3,125
评论 139
转发 245
正文
We trained our Muse foundational models, including Muse Wonder, at Prometheus, our gigawatt-scale supercluster in New Albany, Ohio.
Across a supercluster with hundreds of thousands of GPUs, even one underperforming host can slow an entire training run. We engineered Prometheus for resilience at scale, achieving 85%+ goodput on pretraining, 90%+ on reinforcement learning and ~47K GPU hours mean time between interruptions during Muse Wonder training.
Take a look at the tech powering Prometheus: https://go.meta.me/4yXXApZ