Post-Grokking Collapse at the Representation-Readout Interface in Muon-Trained Transformers

Abstract:Muon-trained modular-arithmetic transformers can lose accuracy while retaining linearly decodable task information. Adjacent swaps localize five captured unnormalized failures to AdamW readout updates. Multiplying the actual readout displacement by the large feature mean produces a class-dependent logit offset shared across inputs that nearly reproduces each failure. Training-only decoders recover 98.20-100% held-out accuracy. Correcting cross-entropy derivative errors stabilizes five matched branches through step 100,000; four prospective accurate-CE RMS runs fail through embedding updates.

38 pages, 11 figures. Revised manuscript with expanded experiments and analysis. Preprint. Under review

Artificial Intelligence (cs.AI); Machine Learning (cs.LG)

arXiv:2608.07436 [cs.AI]

(or arXiv:2608.07436v2 [cs.AI] for this version)

https://doi.org/10.48550/arXiv.2608.07436 Focus to learn more arXiv-issued DOI via DataCite

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.