Metraipolis › archive › dc_debf250906d0
KBBQ: A Predictive Noise Law and the Limits of Spectrum Flattening in FP4 Quantization — arXiv:2609.08135
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.08135
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-16T13:58:08.463Z
- SHA-256 of the archived text
aca98fd9a01b84aac163d778b18943d43398d12117af21da86eda9708d8e6ac5- Stable URL of this page
https://metraipolis.com/documento/dc_debf250906d0
How to cite
KBBQ: A Predictive Noise Law and the Limits of Spectrum Flattening in FP4 Quantization — arXiv:2609.08135 — arXiv (en), https://arxiv.org/abs/2609.08135 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_debf250906d0 (Metraipolis archive dc_debf250906d0, sha256 aca98fd9a01b84aac163d778b18943d43398d12117af21da86eda9708d8e6ac5)
BibTeX
@misc{dcdebf250906d0,
title = {KBBQ: A Predictive Noise Law and the Limits of Spectrum Flattening in FP4 Quantization — arXiv:2609.08135},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.08135},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_debf250906d0; sha256 aca98fd9a01b84aac163d778b18943d43398d12117af21da86eda9708d8e6ac5},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.08135. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
We develop a second-order theory of quantization noise in matrix multiplication in which the quantization format is characterized by the variance it assigns to each element. The constant variance profile of integer quantization recovers existing integer-noise theory, while the multiplicative profile of floating-point rounding reduces the data dependence to a scalar, the participation factor $\kappa$, yielding a closed-form signal-to-noise-ratio law. The resulting functional also admits a closed-form upper bound $\kappa^{*}$ that no function-preserving linear transform can exceed and that is attained by a recent state-of-the-art method. Building on this analysis, we introduce KBBQ (\textbf{K}appa-\textbf{B}raked \textbf{B}lockwise \textbf{Q}uantization), which parameterizes the extent to which a transform approaches this ceiling. At W4A4, across four base models and two FP4 formats, KBBQ outperforms the prior state of the art without additional deployment-time computation.
Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-09 in cs.LG).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.08135 — 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 2026-10-16T13:39:43.023Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-09?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).