arXiv Machine Learning September 30, 2026 By Jonas von Berg, Massimiliano Datres, Carlo Knei{\ss}l, Gitta Kutyniok

Scale Sensitivity in Low-Bit Post-Training Quantization: Curvature of the Quantization Error Landscape

Read the original on arXiv Machine Learning →

The Flow has not summarised this story yet — read it at arXiv Machine Learning.

arXiv AI

Jun 9

ScaleSweep: Accurate NVFP4 Post-Training Quantization of LLMs via Block Scale Initialization

arXiv:2606. 07618v1 Announce Type: cross Abstract: NVFP4 is a recently introduced hardware-supported FP4 format that improves the fidelity of 4-bit quantization through fine-grained block scales.

By Li Lin, Xiaojun Wan

llms efficiency

arXiv Machine Learning

Aug 12

ReRound: Reconstructive Rounding to Resolve Midpoint Ambiguity in Calibration-Free LLM Quantization

arXiv:2608. 11045v1 Announce Type: new Abstract: ReRound (Reconstructive Rounding) is a post-training quantization method that addresses the midpoint ambiguity inherent in standard round-to-nearest (RTN) schemes when quantizing weights near the centers of quantization intervals.

By He-Yen Hsieh, H. T. Kung

llms diffusion efficiency

arXiv AI

Jun 18

HeRo-Q: A General Framework for Stable Low Bit Quantization via Hessian Conditioning

arXiv:2601. 21626v2 Announce Type: replace-cross Abstract: Post Training Quantization (PTQ), a mainstream model compression technique, often leads to the paradoxical 'low error, high loss' phenomenon because it focuses solely on minimizing quantization error.

By Jinhao Zhang, Yunquan Zhang, Zicheng yan, Boyang Zhang, Jun Sun, Daning Cheng

llms efficiency

arXiv Machine Learning

Jun 19

CAGE: Curvature-Aware Gradient Estimation For Accurate Quantization-Aware Training

arXiv:2510. 18784v3 Announce Type: replace Abstract: Despite significant work on low-bit quantization-aware training (QAT), there is still an accuracy gap between such techniques and native training.

By Soroush Tabesh, Mher Safaryan, Andrei Panferov, Alexandra Volkova, Dan Alistarh

llms fine-tuning efficiency benchmarks

arXiv AI

1d ago

Quantization Error Is Spectrally Flat: A Single Random Probe Is a Calibrated, Data-Free Sensitivity Estimator, with Application to Budget-Targeted Mixed-Precision Quantization

arXiv:2609.33923v2 Announce Type: replace-cross Abstract: A single random Gaussian probe gives an unbiased estimate of the squared Frobenius norm of a layer's quantization error. The estimator is wel...

By I Kennedy, T Kennedy

efficiency safety

arXiv AI

Jul 13

Signed Symmetric Quantization for Few-Bit Integers

arXiv:2607. 08779v1 Announce Type: cross Abstract: The signed integer alphabet contains one more negative representable value than positive.

By Ian Colbert, Eashan Dash, Pablo Monteagudo-Lago, Juan Amboage, Srinidhi N, Giuseppe Franco, Nicholas J. Fraser, Arun Ramachandran

llms efficiency

One email a morning, machine-written

Related stories

ScaleSweep: Accurate NVFP4 Post-Training Quantization of LLMs via Block Scale Initialization

ReRound: Reconstructive Rounding to Resolve Midpoint Ambiguity in Calibration-Free LLM Quantization

HeRo-Q: A General Framework for Stable Low Bit Quantization via Hessian Conditioning

CAGE: Curvature-Aware Gradient Estimation For Accurate Quantization-Aware Training

Quantization Error Is Spectrally Flat: A Single Random Probe Is a Calibrated, Data-Free Sensitivity Estimator, with Application to Budget-Targeted Mixed-Precision Quantization

Signed Symmetric Quantization for Few-Bit Integers