Unverified50% confidenceFactExact time
LoRA 由微软提出，其核心思想是将权重更新量 ΔW 分解为两个低秩矩阵的乘积 BA
1
Sources
50%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
UnverifiedThe mathematical principle of LoRA is to decompose the weight update matrix into the product of two low-rank matrices, with file sizes typically ranging from only tens to hundreds of MB.71% similarUnverifiedLoRA在前向传播时原始权重W₀保持冻结，输出变为h = W₀x + (BA)x/√r，对于SD 1.5模型在rank=4时整个LoRA文件约只有3-6MB66% similarUnverifiedLoRA (Low-Rank Adaptation) freezes the full weights of the pre-trained model and only injects two low-rank matrices into specific layers, resulting in file sizes that are typically only a few MB to tens of MB.59% similarVerifiedLoRA (Low-Rank Adaptation) files are typically only tens of MB in size, far smaller than the several GB size of complete models.55% similarUnverifiedLoRA等参数高效微调方法将显存需求压缩至普通消费级GPU可承受的范围55% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/566555API
curl https://kongchang.com/api/v1/knowledge/claims/566555MCP
get_claim(id=566555)