SandwichQuant: Which Parameters Matter Before and After Quantization?

arXiv 2608.24173 ↗First posted 25 Aug 2026cs.CVcited by 0

Not on the map. Why?

Abstract (arXiv)

Quantization correction methods usually optimize weights, quantization parameters, or reconstruction objectives, while the underlying parameter subspaces responsible for effective correction remain unclear. In this work, we study quantization correction from a parameter subspace perspective and reveal that correction capability is highly non-uniform across parameter groups. By decomposing trainable parameters into backbone weights, normalization-affine parameters, and quantization parameters, we show that the low-dimensional normalization-affine subspace provides a highly efficient correction direction under matched budgets. Based on this finding, we propose SandwichQuant, a two-stage normalization-affine correction framework that performs adaptation before and after quantization. The pre-stage improves quantization robustness, while the post-stage compensates residual errors after the quantized graph is fixed. Extensive experiments on vision models and large language models demonstrate consistent improvements under various low-bit quantization settings, validating the effectiveness of subspace-aligned correction.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 51 of its 56 references that are in the map. First posted on arXiv 25 Aug 2026.

Builds on

51 of 56 references.

References · 51

2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 13,028
2015Distilling the Knowledge in a Neural Networkcited by 10,324
2025Qwen3 Technical Reportcited by 8,820
2015Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shiftcited by 7,005
2019Parameter-Efficient Transfer Learning for NLPcited by 4,236
2022GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformerscited by 1,919
2017Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inferencecited by 1,593
2019Root Mean Square Layer Normalizationcited by 1,480
2020Sharpness-Aware Minimization for Efficiently Improving Generalizationcited by 1,417
2022SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Modelscited by 1,386
2016On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minimacited by 1,352
2023AWQ: Activation-aware Weight Quantization for LLM Compression and Accelerationcited by 1,298

38 more

2021BitFit: Simple Parameter-efficient Fine-tuning for Transformer-based Masked Language-modelscited by 1,245
2018Averaging Weights Leads to Wider Optima and Better Generalizationcited by 1,209
2020Intrinsic Dimensionality Explains the Effectiveness of Language Model Fine-Tuningcited by 671
2024QuaRot: Outlier-Free 4-Bit Inference in Rotated LLMscited by 634
2019Learned Step Size Quantizationcited by 628
2020Up or Down? Adaptive Rounding for Post-Training Quantizationcited by 621
2021BRECQ: Pushing the Limit of Post-Training Quantization by Block Reconstructioncited by 479
2024SpinQuant: LLM quantization with learned rotationscited by 471
2017Sharp Minima Can Generalize For Deep Netscited by 466
2018Measuring the Intrinsic Dimension of Objective Landscapescited by 370
2018How Does Batch Normalization Help Optimization?cited by 357
2019Data-Free Quantization Through Weight Equalization and Bias Correctioncited by 336
2021ASAM: Adaptive Sharpness-Aware Minimization for Scale-Invariant Learning of Deep Neural Networkscited by 293
2020LSQ+: Improving low-bit quantization through learnable offsets and better initializationcited by 189
2022QDrop: Randomly Dropping Quantization for Extremely Low-bit Post-Training Quantizationcited by 180
2024SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Modelscited by 170
2024FlatQuant: Flatness Matters for LLM Quantizationcited by 131
2022Overcoming Oscillations in Quantization-Aware Trainingcited by 119
2020Training BatchNorm and Only BatchNorm: On the Expressive Power of Random Features in CNNscited by 104
2025OstQuant: Refining Large Language Model Quantization with Orthogonal and Scaling Transformations for Better Distribution Fittingcited by 97
2020Relative Flatness and Generalizationcited by 88
2022Understanding the Generalization Benefit of Normalization Layers: Sharpness Reductioncited by 86
2025GPTAQ: Efficient Finetuning-Free Quantization for Asymmetric Calibrationcited by 60
2025Quantization Error Propagation: Revisiting Layer-Wise Post-Training Quantizationcited by 41
2021MQBench: Towards Reproducible and Deployable Model Quantization Benchmarkcited by 40
2023Norm Tweaking: High-performance Low-bit Quantization of Large Language Modelscited by 33
2025Qronos: Correcting the Past by Shaping the Future... in Post-Training Quantizationcited by 28
2023Normalization Layers Are All That Sharpness-Aware Minimization Needscited by 27
2024Quantization without Tearscited by 25
2023Fine-tuning Happens in Tiny Subspaces: Exploring Intrinsic Task-specific Subspaces of Pre-trained Language Modelscited by 19
2025APHQ-ViT: Post-Training Quantization with Average Perturbation Hessian Based Reconstruction for Vision Transformerscited by 19
2025FIMA-Q: Post-Training Quantization for Vision Transformers by Fisher Information Matrix Approximationcited by 19
2025DartQuant: Efficient Rotational Distribution Calibration for LLM Quantizationcited by 17
2025Training Dynamics Impact Post-Training Quantization Robustnesscited by 15
2025First-Order Error Matters: Accurate Compensation for Quantized Large Language Modelscited by 8
2026Rethinking Residual Errors in Compensation-based LLM Quantizationcited by 7
2025AIQViT: Architecture-Informed Post-Training Quantization for Vision Transformerscited by 5
2024Robust Training of Neural Networks at Arbitrary Precision and Sparsitycited by 3

50 most cited of 51 All 51 (JSON)

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way