The Active Ingredient in Muon's Grokking

• URL: http://arxiv.org/abs/2607.20512v1
• Date: Mon, 06 Jul 2026 22:35:57 GMT
• Title: The Active Ingredient in Muon's Grokking
• Abstract summary: We show the Muon-faster reaches the grokking threshold on modular arithmetic faster than AdamW.<n>To better understand Moun's behavior, we run multi-seed and multi-learning-rate sweeps to decompose and stress-test the effect.
• Score: 2.921159958223653
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The Muon optimizer reaches the grokking threshold on modular arithmetic faster than AdamW. Prior work attributes this to "spectral-norm constraints plus orthogonalized momentum" but does not isolate which mechanism matters. To better understand Moun's behavior, we run multi-seed and multi-learning-rate sweeps to decompose and stress-test the effect. First, an ablation shows the speedup comes from orthogonalization (the Newton-Schulz iteration): orthogonalize-only matches full Muon, whereas spectral-only is no faster than AdamW and is unreliable, and this verdict holds across learning rates. Second, a mechanistic analysis finds that orthogonalizing optimizers reach generalization at roughly 3x lower spectral norm and, controlling for how much the embedding actually moves, settle into a lower-norm solution rather than simply perturbing the embedding less. Third, reducing the Newton-Schulz iteration count from five to one accelerates reaching the threshold but makes the grokked solution fragile, prone to transient collapse, with fragility that grows with learning rate; a single iteration is fast and stable only at small learning rate, while the canonical five iterations are the learning-rate-robust choice. We also show spectral scaling can be dropped at no measured cost. A methodological thread runs throughout: under a stability-aware metric, "faster" claims about grokking optimizers can invert, so we report both first-crossing and sustained-grok times. To support reproducibility, we release our full training and analysis code at https://github.com/louiswang524/muon-grokking-frontier

Related papers

• REVES: REvision and VErification--Augmented Training for Test-Time Scaling [53.197756110943395]
  We propose a two-stage iterative framework that alternates between online data/prompt augmentation and policy optimization.<n>We observe gains of +6.5 points over the RL baseline and +4.0 points over standard multi-turn training.
  arXiv  Detail & Related papers  (2026-06-17T10:37:23Z)
• Denoise First, Orthogonalize Later: Understanding Momentum in Muon via Spectral Filtering [32.15111956211244]
  We show that momentum in Muon acts as a spectral filter.<n>Applying momentum before perturbation provably stronger alignment with the signal component of gradient than either reversing this order or simply removing momentum.
  arXiv  Detail & Related papers  (2026-06-02T16:54:38Z)
• Spectral Flattening Is All Muon Needs: How Orthogonalization Controls Learning Rate and Convergence [15.754779716816174]
  We show that Muon's maximal stable step size scales with the average singular value of the gradient rather than the largest.<n>We recast Muon as a preconditioned gradient method and show, under a Kronecker-factored curvature model, that it improves the effective convergence factor.
  arXiv  Detail & Related papers  (2026-05-13T06:54:01Z)
• MuonQ: Enhancing Low-Bit Muon Quantization via Directional Fidelity Optimization [7.243820893114047]
  We propose a low-bit Muon training framework built on the principle of directional fidelity optimization.<n>MuonQ at 4-bit precision closely matches full-precision Muon in both training loss and downstream task accuracy.<n>Our code is available at https://github.com/YupSueng/MuonQ.
  arXiv  Detail & Related papers  (2026-05-12T01:31:32Z)
• Accelerating Zeroth-Order Spectral Optimization with Partial Orthogonalization from Power Iteration [6.574641780732972]
  We focus on the hidden layer training problem in which spectrals like Muon outperform AdamW.<n>To do so, we replace the iconic Newton-Schulz procedure in Muon with the faster, more concentrated power-iteration method.<n>Our method can achieve from 1.5x to 4x the convergence speed of ZO-Muon.
  arXiv  Detail & Related papers  (2026-05-09T16:16:45Z)
• Sharp Capacity Scaling of Spectral Optimizers in Learning Associative Memory [77.27772368491698]
  Spectrals such as Muon have recently shown strong empirical performance in large-scale language model training.<n>We study this question through the linear associative memory problem.<n>We show that the storage capacity of Muon significantly exceeds that of SGD.
  arXiv  Detail & Related papers  (2026-03-27T16:13:18Z)
• Muon is Provably Faster with Momentum Variance Reduction [55.388203260208485]
  Recent empirical research has demonstrated that deep learnings based on the linear linear oracle (LMO) over specifically chosen Non-Eudean.<n>Adam-type training methods outperform the minimization of large language models.
  arXiv  Detail & Related papers  (2025-12-18T14:38:39Z)
• MuonBP: Faster Muon via Block-Periodic Orthogonalization [24.232069944820513]
  We show how to adjust the learning rate from the baseline to MuonBP and give guarantees for this algorithm.<n>When training an 8B model with eight-way tensor tensor and ZeRO statewiseing, MuonBP achieves 8% Muon with no degradation in performance.
  arXiv  Detail & Related papers  (2025-10-19T19:56:05Z)
• SKI to go Faster: Accelerating Toeplitz Neural Networks via Asymmetric Kernels [69.47358238222586]
  Toeplitz Neural Networks (TNNs) are a recent sequence model with impressive results. We aim to reduce O(n) computational complexity and O(n) relative positional encoder (RPE) multi-layer perceptron (MLP) and decay bias calls. For bidirectional models, this motivates a sparse plus low-rank Toeplitz matrix decomposition.
  arXiv  Detail & Related papers  (2023-05-15T21:25:35Z)
• Lassoed Tree Boosting [53.56229983630983]
  We prove that a gradient boosted tree algorithm with early stopping faster than $n-1/4$ L2 convergence in the large nonparametric space of cadlag functions of bounded sectional variation. Our convergence proofs are based on a novel, general theorem on early stopping with empirical loss minimizers of nested Donsker classes.
  arXiv  Detail & Related papers  (2022-05-22T00:34:41Z)
• AdamP: Slowing Down the Slowdown for Momentum Optimizers on Scale-invariant Weights [53.8489656709356]
  Normalization techniques are a boon for modern deep learning. It is often overlooked, however, that the additional introduction of momentum results in a rapid reduction in effective step sizes for scale-invariant weights. In this paper, we verify that the widely-adopted combination of the two ingredients lead to the premature decay of effective step sizes and sub-optimal model performances.
  arXiv  Detail & Related papers  (2020-06-15T08:35:15Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.