OrScale: Orthogonalised Optimization with Layer-Wise Trust-Ratio Scaling

• URL: http://arxiv.org/abs/2605.07815v1
• Date: Fri, 08 May 2026 14:47:56 GMT
• Title: OrScale: Orthogonalised Optimization with Layer-Wise Trust-Ratio Scaling
• Abstract summary: We analyze why three natural Muon-LAMB hybrids fail through shape-degenerates, raw-momentum clip saturation, de weight-decay runaway, and show that the decay avoids these failures.<n>We introduce OrScale, a trust-valued layer measure of Muon built on a simple rule.
• Score: 10.699320611591332
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Muon improves neural-network training by orthogonalizing matrix-valued updates, but it leaves each layer's update magnitude controlled mostly by a global learning rate. We introduce OrScale, a trust-ratio extension of Muon built on a simple rule: the denominator of a layer-wise ratio should measure the Frobenius norm of the actual parameter-space direction that will be applied. This yields OrScale for general matrix layers and OrScale-LM for language models, where Moonlight shape scaling is combined with one-time per-layer calibration so every trust ratio starts at one. We analyze why three natural Muon-LAMB hybrids fail through shape-degenerate denominators, raw-momentum clip saturation, and decoupled weight-decay runaway, and show that the real-update-direction denominator with coupled weight decay avoids these failures. Theoretically, OrScale admits an O(1/sqrt(T)) nonconvex convergence guarantee in a nuclear-norm criterion, a strict layer-adaptive descent gain under measurable layer heterogeneity, and calibration properties that preserve muP-style learning-rate transfer at initialization. Empirically, OrScale ranks first on CIFAR-10/DavidNet across three seeds, improving Muon from 93.70% to 94.05% validation top-1, and OrScale-LM improves FineWeb-Edu pre-training versus Muon+Moonlight at three of four scales from 125M to 1.1B parameters while outperforming AdamW at every scale.

Related papers

• Low-Rank Ternary Adaptation for Fine-Tuning Transformers [54.3230438745856]
  Ternary transformers offer extreme memory and compute efficiency.<n>Existing low-bit LoRA-based methods cannot directly fine-tune ternary weights.<n>We propose ternary multiplicative adaptation, which represents discrete updates of ternary weights.
  arXiv  Detail & Related papers  (2026-08-25T12:15:56Z)
• Hidden Decoding at Scale: Latent Computation Scaling for Large Language Models [75.80275843320511]
  We study whether an existing backbone can keep improving by allocating more computation to each token while leaving the Transformer backbone fixed.<n>We propose Hidden Decoding, a sequence-length scaling method applied during continued pretraining.
  arXiv  Detail & Related papers  (2026-07-09T07:37:59Z)
• Improving Neural Network Training by Decoupling the Magnitude and Direction of Weight Vectors [110.10775872372047]
  We propose emphMagnitude-Direction-Direction (MD) Decoupling.<n>It factorizes each weight into a fixed-norm direction on a hypersphere and learnable per-row and per-column magnitude gains, updated at separate learning rates.<n>Across both Adam and Muon, MD Decoupling improves on well-tuned baselines, transfers the optimal LR across model width without retuning, and continues to help at scale on large Mixture-of-Experts models.
  arXiv  Detail & Related papers  (2026-06-24T15:40:26Z)
• cMoLLM at Scale: Horizontal Scaling Laws for Mixture-of-LLMs [3.50654760771471]
  cMoLLM is a convolutionally gated mixture-of-LLMs that routes over end-to-end streams through fully differentiable dynamic convolution.<n>In GPT-2-style models trained on FineWeb, cMoLLM improves language modeling perplexity and downstream GLUE and SQuAD accuracy under matched compute.
  arXiv  Detail & Related papers  (2026-06-06T13:51:15Z)
• Distance-Aware Muon: Adaptive Step Scaling for Normalized Optimization [47.817303172845506]
  Muon and related normalizeds decouple the choice of update direction from the choice of step scale.<n>We develop three complementary algorithms for Muon tuning and develop the normalized adaptive scaling parameter.
  arXiv  Detail & Related papers  (2026-05-18T18:19:26Z)
• Scalable Adaptation of 3D Geometric Foundation Models via Weak Supervision from Internet Video [76.32954467706581]
  We propose SAGE, a framework for Scalable Adaptation of GEometric foundation models from raw video streams.<n>We use a hierarchical mining pipeline to transform videos into training trajectories and hybrid supervision.<n>Experiments show that SAGE significantly enhances zero-shot generalization, reducing Chamfer Distance by 20-42% on unseen benchmarks.
  arXiv  Detail & Related papers  (2026-02-08T09:53:21Z)
• MLPMoE: Zero-Shot Architectural Metamorphosis of Dense LLM MLPs into Static Mixture-of-Experts [0.0]
  Large Language Models (LLMs) are predominantly deployed as dense transformers, where every parameter in every feed-forward block is activated for every token.<n>Recent upcycling methods such as MoEfication, CMoE, ToMoE, and MoORE reveal that much of the useful computation lives in sparse, semi-modular substructures inside dense feed-forward networks.<n>This paper introducesMoE (MLP-Experts), a training-free transformation that restructures the dense in transformer blocks into a static, high-cardinality mixture of experts.
  arXiv  Detail & Related papers  (2025-11-26T06:14:26Z)
• Diffusion Language Models are Super Data Learners [61.721441061210896]
  When unique data is limited, diffusion language models (DLMs) consistently surpass autoregressive (AR) models by training for more epochs.<n>We attribute the gains to three compounding factors: (1) any-order modeling, (2) super-dense compute from iterative bidirectional denoising, and (3) built-in Monte Carlo augmentation.
  arXiv  Detail & Related papers  (2025-11-05T08:17:42Z)
• 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)
• Robust Layerwise Scaling Rules by Proper Weight Decay Tuning [50.11170157029911]
  In modern scale-invariant architectures, training quickly enters an degrading-governed steady state.<n>We introduce a weight-decay scaling rule for AdamW that preserves sublayer gain across widths.<n>Our results extend $mu$P beyond the near-init regime by explicitly controlling the steady-state scales set by parameters.
  arXiv  Detail & Related papers  (2025-10-17T02:58:35Z)
• Normalization Layer Per-Example Gradients are Sufficient to Predict Gradient Noise Scale in Transformers [2.1415873597974286]
  Per-example gradient norms are a vital ingredient for estimating gradient noise scale (GNS) with minimal variance. We propose a method with minimal FLOPs in 3D or greater tensor regimes by simultaneously computing the norms while computing the parameter gradients. We find that the total GNS of contemporary transformer models is predicted well by the GNS of only the normalization layers.
  arXiv  Detail & Related papers  (2024-11-01T19:50:00Z)
• Zeroth-Order Fine-Tuning of LLMs in Random Subspaces [63.10833446782114]
  As language models grow in size, memory demands for backpropagation increase.<n>Zeroth-order (ZO) optimization methods offer a memory-efficient alternative.<n>In this paper, we propose Subspace Zero-order optimization to address the challenges posed by posed by high dimensionality perturbations.
  arXiv  Detail & Related papers  (2024-10-11T17:01:43Z)
• Back to MLP: A Simple Baseline for Human Motion Prediction [59.18776744541904]
  This paper tackles the problem of human motion prediction, consisting in forecasting future body poses from historically observed sequences. We show that the performance of these approaches can be surpassed by a light-weight and purely architectural architecture with only 0.14M parameters. An exhaustive evaluation on Human3.6M, AMASS and 3DPW datasets shows that our method, which we dub siMLPe, consistently outperforms all other approaches.
  arXiv  Detail & Related papers  (2022-07-04T16:35:58Z)
• Robust Training of Neural Networks using Scale Invariant Architectures [70.67803417918854]
  In contrast to SGD, adaptive gradient methods like Adam allow robust training of modern deep networks. We show that this general approach is robust to rescaling of parameter and loss. We design a scale invariant version of BERT, called SIBERT, which when trained simply by vanilla SGD achieves performance comparable to BERT trained by adaptive methods like Adam.
  arXiv  Detail & Related papers  (2022-02-02T11:58:56Z)
• NormFormer: Improved Transformer Pretraining with Extra Normalization [31.250781670447257]
  NormFormer architecture adds three normalization operations to each layer. Extra operations incur negligible compute cost. For masked language modeling, NormFormer improves fine-tuned GLUE performance by 1.9%.
  arXiv  Detail & Related papers  (2021-10-18T16:47:45Z)

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.