Metraipolis › archive › dc_951cefc702d6
CurveFP: Co-Designing Numerical Representation and Product Arithmetic for Language Models — arXiv:2608.10010
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.10010
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-19T06:19:13.259Z
- SHA-256 of the archived text
4b996a7371aba213f3a149452fd984a43fbd851bbcb009bb611f5e7123cf27c9- Stable URL of this page
https://metraipolis.com/documento/dc_951cefc702d6
How to cite
CurveFP: Co-Designing Numerical Representation and Product Arithmetic for Language Models — arXiv:2608.10010 — arXiv (en), https://arxiv.org/abs/2608.10010 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_951cefc702d6 (Metraipolis archive dc_951cefc702d6, sha256 4b996a7371aba213f3a149452fd984a43fbd851bbcb009bb611f5e7123cf27c9)
BibTeX
@misc{dc951cefc702d6,
title = {CurveFP: Co-Designing Numerical Representation and Product Arithmetic for Language Models — arXiv:2608.10010},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.10010},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_951cefc702d6; sha256 4b996a7371aba213f3a149452fd984a43fbd851bbcb009bb611f5e7123cf27c9},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.10010. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Low-precision formats usually optimize scalar fidelity while inheriting conventional product arithmetic. We introduce CurveFP, a block-scaled family that distributes magnitudes across interleaved logarithmic curves. Uniform curve indices make every nonzero product an exact sign and integer-index update, while a rational radix exposes the finite phase schedule required for accumulation. We instantiate the algebra as CurveFP8 E4C3/E5C2 for training and CurveFP7 E3C3 for compact inference. On four 7B-9B models, CurveFP7 beats tensorwise FP8 perplexity with one fewer element bit and stays within 1.32% of native quality. CurveFP8 lowers error in all 36 paired training-GEMM comparisons. Across three matched 3B-token pretraining triplets, it reaches mean BF16-inference perplexity 22.5366 versus 22.5407 for FP8 and has a lower format penalty in every seed. Downstream evaluation shows transfer parity and a consistent WikiText-103 gain. In a preliminary 4x4 Nangate45 spatial accelerator tile, CurveFP8 uses one fewer product register and 4.6% less area than timing-closing FP8 at 500 MHz. These results support CurveFP as a numerical and arithmetic co-design, while leaving system-level efficiency to future study. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-12 in cs.LG). Categorie arXiv: Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2608.10010 — 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 2026-10-19T06:11:45.060Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-08-12?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).