Normalized Architectures are Natively 4-Bit

• URL: http://arxiv.org/abs/2605.06067v1
• Date: Thu, 07 May 2026 11:54:07 GMT
• Title: Normalized Architectures are Natively 4-Bit
• Abstract summary: nGPT, an architecture that constrains weights and hidden representations to the unit hypersphere, is inherently more robust to low-precision arithmetic.<n>We validate this approach on both a 1.2B dense model and hybrid (Mamba-Transformer) MoE models of up to 3B/30B parameters.
• Score: 49.13186675123547
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Training large language models at 4-bit precision is critical for efficiency. We show that nGPT, an architecture that constrains weights and hidden representations to the unit hypersphere, is inherently more robust to low-precision arithmetic. This removes the need for interventions-such as applying random Hadamard transforms and performing per-tensor scaling calculations-to preserve model quality, and it enables stable end-to-end NVFP4 training. We validate this approach on both a 1.2B dense model and hybrid (Mamba-Transformer) MoE models of up to 3B/30B parameters. We trace this robustness to the dot product: while quantization noise remains largely uncorrelated in both standard and normalized architectures, the signal behaves differently. In nGPT, the hypersphere constraint enhances weak positive correlations among the element-wise products, leading to a constructive accumulation of the signal across the hidden dimension while the noise continues to average out. This yields a higher effective signal-to-noise ratio and a flatter loss landscape, with the effect strengthening as the hidden dimension grows, suggesting increasing advantages at scale. A reference implementation is available at https://github.com/anonymous452026/ngpt-nvfp4

Related papers

• Architecture-Dependent Causal Transfer of Activation States Across Large Language Models [51.56484100374058]
  Internal language activation states can be transferred causally between different large language models.<n>Injecting projected activations into a target model during generation produces a statistically significant, preregistered causal effect.<n>We conclude that end-to-end activation-state transfer between LLMs, as currently implemented, is architecture-dependent rather than universal.
  arXiv  Detail & Related papers  (2026-08-17T09:53:27Z)
• Precision Tracked Transformer via Kalman Filtering, Kriging and Process Noise [11.579218102536762]
  The Transformer is the foundational building block of modern AI, yet offers no principled handling of emphuncertainty<n>We show this is a case of our emphBayesian Filtering Transformer (BFT)<n>BFT replaces any Transformer layer with negligible overhead.
  arXiv  Detail & Related papers  (2026-05-12T22:22:15Z)
• BWTA: Accurate and Efficient Binarized Transformer by Algorithm-Hardware Co-design [71.97035034203275]
  We analyze zero-point distortion in binarization and propose a Binary Weights & Ternary Activations (BWTA) quantization scheme.<n>For training, we propose Smooth Multi-Stage Quantization, combining a Levelwise Degradation Strategy and a MagnitudeAlignment Projection Factor.<n>Experiments show that BWTA approaches full-precision performance for Transformer-based models, with an average 3.5% drop on GLUE and less than 2% drop on five tasks.
  arXiv  Detail & Related papers  (2026-04-05T04:25:07Z)
• NoiseFormer -- Noise Diffused Symmetric Attention Transformer [0.0]
  We propose a novel unified model architecture called Noise Diffused Symmetric Attention Transformer to enhance the model's performance.<n>The proposed model is validated upon GPT2 base model and the results reflect the performance gains falling between plain Symmetric attention and GPT2 base model.
  arXiv  Detail & Related papers  (2026-01-10T14:10:48Z)
• Unifying Sign and Magnitude for Optimizing Deep Vision Networks via ThermoLion [0.0]
  Current paradigms impose a static compromise on information channel drift parameters.<n>We introduce a "low-dimensional" exploration model and a "low-dimensional" dynamic alignment framework.
  arXiv  Detail & Related papers  (2025-12-01T17:04:17Z)
• Optimized Minimal 4D Gaussian Splatting [50.29519144753797]
  4D Gaussian Splatting has emerged as a new paradigm for dynamic scene representation, enabling real-time rendering of scenes with complex motions.<n>It faces a major challenge of storage overhead, as millions of Gaussians are required for high-fidelity reconstruction.<n>We present OMG4, a framework that constructs a compact set of salient Gaussians capable of faithfully representing 4D Gaussian models.
  arXiv  Detail & Related papers  (2025-10-04T16:11:13Z)
• Precise Bayesian Neural Networks [0.0]
  We develop a lightweight, implementation-ready variational unit that fits modern normalized architectures and improves calibration without sacrificing accuracy.<n>In short, by aligning the variational posterior with the network's intrinsic geometry, BNNs can be simultaneously principled, practical, and precise.
  arXiv  Detail & Related papers  (2025-06-24T15:42:00Z)
• Pushing the Limits of Low-Bit Optimizers: A Focus on EMA Dynamics [64.62231094774211]
  Statefuls (e.g., Adam) maintain auxiliary information even 2x the model size in order to achieve optimal convergence.<n>SOLO enables Adam-styles to maintain quantized states with precision as low as 3 bits, or even 2 bits.<n>SOLO can thus be seamlessly applied to Adam-styles, leading to substantial memory savings with minimal accuracy loss.
  arXiv  Detail & Related papers  (2025-05-01T06:47:45Z)
• Hyperspherical Normalization for Scalable Deep Reinforcement Learning [57.016639036237315]
  SimbaV2 is a novel reinforcement learning architecture designed to stabilize optimization.<n>It scales up effectively with larger models and greater compute, achieving state-of-the-art performance on 57 continuous control tasks.
  arXiv  Detail & Related papers  (2025-02-21T08:17:24Z)
• The random coupled-plaquette gauge model and the surface code under circuit-level noise [1.351813974961217]
  We optimally account for genuine Y-errors in the surface code in a setting with noisy measurements.<n>We tackle the circuit-level noise scenario, where we use a reduction technique to find effective asymmetric depolarizing and syndrome noise rates.
  arXiv  Detail & Related papers  (2024-12-18T16:20:14Z)
• SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Models [61.474101404805545]
  Diffusion models can generate high-quality images, but as they scale, rising memory demands and higher latency pose deployment challenges.<n>We propose SVDQuant, a new 4-bit quantization paradigm to overcome this limitation.<n>We reduce the memory usage for the 12B FLUX.1 models by 3.5$times$, achieving 3.0$times$ speedup over the 4-bit weight-only quantization (W4A16) baseline.
  arXiv  Detail & Related papers  (2024-11-07T18:59:58Z)
• Adaptive Sparsity Level during Training for Efficient Time Series Forecasting with Transformers [20.23085795744602]
  We propose textbfAdaptive textbfSparsity textbfLevel (textbfPALS) to automatically seek a decent balance between loss and sparsity. PALS draws inspiration from sparse training and during-training methods. It introduces the novel "expand" mechanism in training sparse neural networks, allowing the model to dynamically shrink, expand, or remain stable to find a proper sparsity level.
  arXiv  Detail & Related papers  (2023-05-28T06:57:27Z)
• Exact Backpropagation in Binary Weighted Networks with Group Weight Transformations [0.0]
  Quantization based model compression serves as high performing and fast approach for inference. Models that constrain the weights to binary values enable efficient implementation of the ubiquitous dot product.
  arXiv  Detail & Related papers  (2021-07-03T10:29:34Z)

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.