Metraipolis › archive › dc_4871ec90b82d

StagQ: Constraint-Driven Multi-Precision Weight Quantization for LLMs — arXiv:2610.05977

Source
arXiv
Original
https://arxiv.org/abs/2610.05977
License (as archived)
CC0-1.0
Archived
2026-10-06T05:28:53.675Z
SHA-256 of the archived text
a6fdce5fee09b1f94fdf3d61e9df6fa38700c5a611c3093bb56c1f0c738cb247
Stable URL of this page
https://metraipolis.com/documento/dc_4871ec90b82d

How to cite

StagQ: Constraint-Driven Multi-Precision Weight Quantization for LLMs — arXiv:2610.05977 — arXiv, https://arxiv.org/abs/2610.05977 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_4871ec90b82d (Metraipolis archive dc_4871ec90b82d, sha256 a6fdce5fee09b1f94fdf3d61e9df6fa38700c5a611c3093bb56c1f0c738cb247)
BibTeX
@misc{dc4871ec90b82d,
  title = {StagQ: Constraint-Driven Multi-Precision Weight Quantization for LLMs — arXiv:2610.05977},
  howpublished = {arXiv},
  url = {https://arxiv.org/abs/2610.05977},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_4871ec90b82d; sha256 a6fdce5fee09b1f94fdf3d61e9df6fa38700c5a611c3093bb56c1f0c738cb247},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv, https://arxiv.org/abs/2610.05977. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Serving a large language model (LLM) across a fleet of deployments requires several weight-precision operating points. Multi-precision formats serve them all from one stream whose prefixes are valid lower-precision codes, instead of storing multiple copies. We present StagQ, a multi-precision weight format whose main stream is a 2-bit group-wise affine base followed by a configurable number of 1-bit refinement planes on a dyadic step schedule. Every supported precision is a readable prefix, decoded by an affine map derived from metadata shared across all precisions, with no per-weight lookup. A sparse side record, filled both before and after the grid is fitted, holds out the few weights the grid serves worst. We report two configurations of the encoder. At two bits the cheaper one leads the strongest multi-precision baseline on Llama-3.1-8B, Phi-4, and OLMo-2-7B by 3.1 to 7.0 MMLU points, at a slightly lower logical rate. At three bits it leads on Llama-3.1-8B, leads on Phi-4 at a higher rate, and ties on OLMo-2-7B. At four bits it ties on all three, at a higher rate. In a batch-one matrix-vector product on an NVIDIA A100 GPU, timed on synthetic weights, our kernel is faster than the two baseline kernels in most shape-precision cases.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2610.05977 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-06T05:24:20.453Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).