Research

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

Single source1 source7 Oct 2026

arXiv:2608.07436v2 Announce Type: replace Abstract: Muon-trained modular-arithmetic transformers can lose accuracy while retaining linearly decodable task information.

Reported by 1 source

arxiv.orghttps://arxiv.org/abs/2608.07436

← Back to all developments · Signals are a development reported by more than one source. See the methodology.