---
id: 20260813-T0-08
title: "CurveFP：专为大模型设计的新型低精度数据格式，提升推理速度"
title_en: "CurveFP: New Low-Precision Datatype Boosts LLM Inference Speed"
url: https://ai.daily.yangsir.net/daily/20260813-T0-08
issue_date: 2026-08-13
publish_date: 2026-08-12T04:00:00.000Z
category: research
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2608.10010
---

# CurveFP：专为大模型设计的新型低精度数据格式，提升推理速度

一篇来自arXiv的论文提出了CurveFP，一种基于有理根（rational-radix）对数的封闭乘积码本（closed-product codebook）数据类型。与现有低精度格式不同，CurveFP在量化幅度的同时，优化了乘积运算的闭包性，使得算术操作在低精度下也能保持较高精度。在实际应用中，CurveFP能有效降低大语言模型的计算和存储成本，在不显著损失精度的前提下，提升推理速度。该方案为AI硬件和模型部署领域的开发者提供了一种新的低精度方案选择。

## English Version

**CurveFP: New Low-Precision Datatype Boosts LLM Inference Speed**

A paper on arXiv introduces CurveFP, a rational-radix logarithmic datatype with closed-product codebook designed for language models. Unlike existing low-precision formats, CurveFP quantizes magnitudes while ensuring the closure of product operations, maintaining high arithmetic accuracy even at low precisions. The approach reduces computational and storage costs for large language models, enabling faster inference without significant accuracy loss. This offers a new low-precision option for AI hardware and model deployment.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2608.10010)

**详情页**：https://ai.daily.yangsir.net/daily/20260813-T0-08

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*