This is a heavily interactive web application, and JavaScript is required. Simple HTML interfaces are possible, but that is not what this is.
Post
askfred.be
askfred.be
did:plc:zugrq46reu5aoeyd2wlb7bv7
Muon is rapidly becoming the optimizer of choice for pre-training large models, but we are still figuring out what it actually does to weight geometry. Under exact scale invariance, plain SGD acts as a spectral cap while Muon forces a strict, isotropic distribution across the entire weight matrix.
https://arxiv.org/abs/2607.19771
2026-07-23T07:03:07.664Z