CoreQ: Learning-Free Mismatch Correction and Successive Rounding for Quantization

TL;DR

CoreQ is a learning-free post-training quantization framework that adaptively corrects layer mismatch errors in large language models, improving performance without overfitting or hyperparameter tuning.

Contribution

It introduces a closed-form mismatch correction coefficient based on geometric decomposition, enabling adaptive, hyperparameter-free calibration for quantized models.

Findings

CoreQ outperforms strong PTQ baselines across multiple LLMs and quantization settings.

The method reduces overfitting by avoiding fixed mismatch correction scaling.

CoreQ improves perplexity and downstream accuracy in experiments.

Abstract

Post-training quantization (PTQ) enables efficient deployment of large language models by mapping pretrained weights to low-bit formats without retraining, typically using a small calibration set to minimize a layer-wise calibration objective. However, this sequential procedure induces a mismatch: errors from earlier quantized layers alter the inputs received by later layers, causing the activations to deviate from those of the full-precision model. Recent approaches introduce mismatch-aware calibration objectives to compensate for this effect, but leave open how much of the observed mismatch should shift each layer's calibration target. Fully applying this correction can overfit limited calibration data, while scaling the mismatch correction with a fixed coefficient ignores varying reliability of mismatch estimates across layers. To address these limitations, we propose CoreQ, a learning-free PTQ framework that applies a closed-form coefficient for mismatch correction derived from a geometric decomposition of the mismatch. The resulting coefficient adapts the correction across layers, reduces overfitting to finite calibration data, and requires no hyperparameter tuning. Given the corrected target, CoreQ minimizes the induced triangular least-squares objective with an efficient greedy successive-rounding solver and a bounded beam-search extension, K-CoreQ, that trades modest additional compute for improved performance. Across multiple LLM families, scales, bit-widths, and quantization settings, CoreQ improves perplexity and downstream accuracy over strong PTQ baselines.