Unverified60% confidenceFactExact time

LoRA将微调权重变化量ΔW分解为两个低秩矩阵的乘积，参数量仅为原模型的0.1%~1%，当r=8、d=4096时参数压缩比高达256倍

2

Sources

60%

Confidence

Long-term

Relevance

7/14/2026

First Seen

Sources

零基础学AI大模型：避开三大误区的完整学习路径

Related Claims

VerifiedLoRA中低秩矩阵的秩r通常设为8或16，可训练参数量从d×k降至r×(d+k)，降幅可达数百倍74% similar UnverifiedLoRA由微软研究院于2021年提出，对权重更新量ΔW进行低秩分解ΔW = A·B，参数量通常可压缩至原来的1/100以下73% similar UnverifiedLoRA将可训练参数量从d²降低到2dr，其中r通常为4-6469% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/504219

API

curl https://kongchang.com/api/v1/knowledge/claims/504219

MCP

get_claim(id=504219)