Pith. sign in

REVIEW 17 cited by

nGPT: Normalized Transformer with Representation Learning on the Hypersphere

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2410.01131 v2 pith:MZTO45HE submitted 2024-10-01 cs.LG cs.AI

nGPT: Normalized Transformer with Representation Learning on the Hypersphere

classification cs.LG cs.AI
keywords hyperspherengptnormalizedattentionlearningrepresentationsametransformer
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

We propose a novel neural network architecture, the normalized Transformer (nGPT) with representation learning on the hypersphere. In nGPT, all vectors forming the embeddings, MLP, attention matrices and hidden states are unit norm normalized. The input stream of tokens travels on the surface of a hypersphere, with each layer contributing a displacement towards the target output predictions. These displacements are defined by the MLP and attention blocks, whose vector components also reside on the same hypersphere. Experiments show that nGPT learns much faster, reducing the number of training steps required to achieve the same accuracy by a factor of 4 to 20, depending on the sequence length.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 17 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Size Doesn't Matter: Cosine-Scored Sparse Autoencoders

    cs.LG 2026-06 unverdicted novelty 7.0

    Cosine-scored SAEs with a learned direction-magnitude blend learn more concept-aligned features than standard inner-product SAEs at matched reconstruction quality.

  2. Chem-GMNet: A Sphere-Native Geometric Transformer for Molecular Property Prediction

    cs.LG 2026-05 unverdicted novelty 7.0

    Chem-GMNet uses sphere-native embeddings, DualSKA attention, and SH-FFN layers to match or beat ChemBERTa-2 on MoleculeNet tasks with fewer parameters and sometimes no pretraining.

  3. Universal One-third Time Scaling in Learning Peaked Distributions

    cs.LG 2026-02 conditional novelty 7.0

    Softmax + cross-entropy on peaked targets yields loss ∼ t^{−1/3}, giving an architecture-driven explanation for power-law LLM training time without power-law data.

  4. Robust Filter Attention: Self-Attention as Precision-Weighted State Estimation

    cs.LG 2025-09 unverdicted novelty 7.0

    Robust Filter Attention models self-attention as consistency-based state estimation under a linear SDE for token trajectories, matching standard attention complexity while showing lower perplexity and better zero-shot...

  5. Improving Neural Network Training by Decoupling the Magnitude and Direction of Weight Vectors

    cs.LG 2026-06 unverdicted novelty 6.0

    MD Decoupling factorizes weights into fixed-norm directions and learnable per-row/column magnitudes updated at independent rates, improving Adam and Muon training stability and scale transfer without weight decay or warmup.

  6. The Structural Origin of Attention Sink: Variance Discrepancy, Super Neurons, and Dimension Disparity

    cs.LG 2026-05 unverdicted novelty 6.0

    Attention sinks arise from variance discrepancy in self-attention value aggregation, amplified by super neurons and first-token dimension disparity, and can be mitigated by head-wise RMSNorm to accelerate pre-training...

  7. Demystifying Manifold Constraints in LLM Pre-training

    cs.LG 2026-05 unverdicted novelty 6.0

    Manifold constraints via the new MACRO optimizer independently bound activation scales and enforce rotational equilibrium in LLM pre-training, subsuming RMS normalization and decoupled weight decay while delivering co...

  8. Polaris: Coupled Orbital Polar Embeddings for Hierarchical Concept Learning

    cs.LG 2026-04 unverdicted novelty 6.0

    Polaris separates semantic meaning from hierarchical structure in embeddings via angular geometry and radius on a hypersphere, yielding up to 19-point gains in taxonomy expansion retrieval over baselines.

  9. Polaris: Coupled Orbital Polar Embeddings for Hierarchical Concept Learning

    cs.LG 2026-04 unverdicted novelty 6.0

    Polaris learns hierarchical concepts via coupled orbital polar embeddings on hyperspheres that separate meaning from structure using tangent projections, exponential maps, and asymmetric objectives, yielding up to 19-...

  10. FlashSAC: Fast and Stable Off-Policy Reinforcement Learning for High-Dimensional Robot Control

    cs.LG 2026-04 unverdicted novelty 6.0

    FlashSAC scales up Soft Actor-Critic with fewer updates, larger models, higher data throughput, and norm bounds to deliver faster, more stable training than PPO on high-dimensional robot control tasks across dozens of...

  11. FlashSAC: Fast and Stable Off-Policy Reinforcement Learning for High-Dimensional Robot Control

    cs.LG 2026-04 unverdicted novelty 6.0

    FlashSAC improves training speed and final performance of off-policy RL on high-dimensional robot tasks by reducing update frequency, increasing model scale, and bounding norms to limit critic error accumulation.

  12. Robust Filter Attention: Self-Attention as Precision-Weighted State Estimation

    cs.LG 2025-09 reject novelty 6.0

    AFA re-derives self-attention as a precision-weighted robust state estimator for a linear SDE, recovering rotary encodings in a limit, but the abstract's language-modeling results are absent from the text.

  13. Superposition Yields Robust Neural Scaling

    cs.LG 2025-05 conditional novelty 6.0

    Strong superposition causes neural loss to scale as the inverse of model dimension due to geometric feature overlaps, explaining scaling laws for broad frequency distributions.

  14. Normalized Matching Transformer

    cs.CV 2025-03 unverdicted novelty 6.0

    Normalized Matching Transformer enforces unit-norm embeddings at every Transformer layer and trains with InfoNCE plus hyperspherical uniformity loss, reaching new state-of-the-art accuracy on PascalVOC and SPair-71k w...

  15. Improving Neural Network Training by Decoupling the Magnitude and Direction of Weight Vectors

    cs.LG 2026-06 conditional novelty 5.0

    Splitting weight matrices into a fixed-norm direction and learnable per-row/column magnitudes improves LLM training over AdamW/Muon, removes weight decay and warmup, and transfers the optimal LR across width.

  16. Constrained Stochastic Spectral Preconditioning Converges for Nonconvex Objectives

    math.OC 2026-05 unverdicted novelty 5.0

    Proximal stochastic spectral preconditioning converges for nonconvex constrained objectives under heavy-tailed noise, with a variance-reduced version achieving faster rates and a refined analysis of Muon iterations.

  17. When Does Removing LayerNorm Help? Activation Bounding as a Regime-Dependent Implicit Regularizer

    cs.LG 2026-04 unverdicted novelty 5.0

    DyT improves validation loss 27% at 64M params/1M tokens but worsens it 19% at 118M tokens, with saturation levels predicting the sign of the effect.