Constraining weight matrices to a manifold makes update sizes interpretable, prevents norm blowup, bounds condition number, and supports Lipschitz guarantees. Muon can be co-designed with that constraint rather than bolted on.

• articleTraining Llms To Predict World Events
• articleLearning To Replicate Expert Judgment In Financial Tasks
• articleAnnouncing Tinker
• articleTinker General Availability

• videoCompression at the Edge — Chris Alexiuk, NVIDIAAI Engineer
• clipThe bet: under 1% of rollout-visible weights change per versionAI Engineer
• clipOpen questions: Muon, fully async RL, and other training stagesAI Engineer

Checking sign-in…

Loading comments…