{
  "abstract": "Orthogonal transforms have emerged as an effective tool for low-bit quantization of large language models (LLMs), as they reshape activation distributions without changing the underlying model function. Existing approaches typically rely either on fixed data-agnostic transforms, such as Hadamard rotations, or on calibration-time optimization of learned rotations. We study rotation selection from first principles and show that maximizing covariance leads to spectral alignment and provides a principled criterion for constructing quantization-friendly orthogonal transforms. Based on this insight, we propose SpectrumQuant, a closed-form rotation method that estimates activation covariance from calibration data and uses it to construct a covariance-aware orthogonal transform for post-training quantization. Our formulation characterizes quantization-friendly rotations as compositions of alignment and mixing components, while recovering Hadamard matrices as a special case. Interestingly, a class of butterfly matrices belongs to this category. In the realm of LLM quantization, SpectrumQuant bridges the gap between data-agnostic structured rotations and learned rotations, providing an effective and efficient data-aware alternative for post-training quantization. Experiments on Llama and Mistral models show that SpectrumQuant is competitive with state-of-the-art rotation strategies while requiring less than one-third of the GPU memory and significantly less time than learned rotations approaches like SpinQuant and DartQuant.",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": null,
  "arxiv": "2610.20303",
  "author_total": 4,
  "authors": [
    {
      "name": "Dwaipayan Haldar"
    },
    {
      "name": "Indranil Patra"
    },
    {
      "name": "Azhar Yousuf"
    },
    {
      "name": "Chandra Sekhar Seelamantula"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/4hmnzysss5tmn.bib",
  "cited_by": 0,
  "doi": "10.48550/arxiv.2610.20303",
  "fulltext_url": null,
  "id": "4hmnzysss5tmn",
  "issns": [],
  "landing_url": "https://arxiv.org/abs/2610.20303",
  "language": null,
  "licence": "LicenseRef-arXiv-nonexclusive-1.0",
  "licence_label": "arXiv licence",
  "licence_source": "article",
  "licence_url": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
  "metadata_licence": "CC0-1.0",
  "native_subjects": [
    [
      "arxiv",
      "cs.LG"
    ],
    [
      "arxiv",
      "stat.ML"
    ],
    [
      "openalex",
      "1707|Advanced Neural Network Applications"
    ]
  ],
  "openalex": "W7222212734",
  "pdf_url": "https://arxiv.org/pdf/2610.20303",
  "pmcid": null,
  "pmid": null,
  "published": "2026-10-23",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/4hmnzysss5tmn.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T12:35:25Z",
      "native_id": "2610.20303",
      "source": "arxiv",
      "url": "https://arxiv.org/abs/2610.20303"
    }
  ],
  "subjects": [
    {
      "field": "mathematics",
      "origin": "arxiv:stat.ML",
      "subfield": "statistics"
    }
  ],
  "title": "SpectrumQuant: Spectrum Alignment for LLM Quantization",
  "title_text": "SpectrumQuant: Spectrum Alignment for LLM Quantization",
  "type": "preprint",
  "updated_at": "2026-10-30T12:35:25Z",
  "url": "https://research.draconitic.com/paper/4hmnzysss5tmn",
  "venue": "arXiv",
  "versions": []
}