[2603.22324] DAQ: Delta-Aware Quantization for Post-Training LLM Weight Compression

Computer Science > Machine Learning arXiv:2603.22324 (cs) [Submitted on 20 Mar 2026] Title:DAQ: Delta-Aware Quantization for Post-Training LLM Weight Compression Authors:Xiaoming Yu, Shize Tang, Guanghua Yu, Linchuan Xie, Song Liu, Jianchen Zhu, Feng Li View a PDF of the paper titled DAQ: Delta-Aware Quantization for Post-Training LLM Weight Compression, by Xiaoming Yu and 6 other authors View PDF HTML (experimental) Abstract:We introduce Delta-Aware Quantization (DAQ), a data-free post-training quantization framework that preserves the knowledge acquired during post-training. Standard quantization objectives minimize reconstruction error but are agnostic to the base model, allowing quantization noise to disproportionately corrupt the small-magnitude parameter deltas ($\Delta W$) that encode post-training behavior -- an effect we analyze through the lens of quantization as implicit regularization. DAQ replaces reconstruction-based objectives with two delta-aware metrics -- Sign Preservation Rate and Cosine Similarity -- that directly optimize for directional fidelity of $\Delta W$, requiring only the base and post-trained weight matrices. In a pilot FP8 study, DAQ recovers style-specific capabilities lost under standard quantization while maintaining general performance. Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI) Cite as: arXiv:2603.22324 [cs.LG]   (or arXiv:2603.22324v1 [cs.LG] for this version)   https://doi.org/10.48550/arXiv.2603.22324 Focus to...