Reaction
Muse Wonder was trained at Meta’s Prometheus supercluster in Ohio
Meta says Prometheus reached 85%-plus pretraining goodput and 90%-plus reinforcement-learning goodput during Muse Wonder training.
TLDR
Meta says it trained its Muse foundational models, including Muse Wonder, at its gigawatt-scale Prometheus supercluster in New Albany, Ohio. During Muse Wonder training, Meta reports 85%-plus goodput on pretraining, 90%-plus on reinforcement learning and roughly 47,000 GPU hours mean time between interruptions.
Combined views
8.8K
3 Sources, first seen ago
