Unverified60% confidenceFactExact time
LoRA was proposed by Hu et al. at Microsoft Research in 2021; by decomposing the weight update matrix into the product of two low-rank matrices, it reduces GPU memory usage by approximately 3×.
2
Sources
60%
Confidence
Long-term
Relevance
7/24/2026
First Seen
Sources
Related Claims
UnverifiedLoRA was proposed by Microsoft Research in 2021; it injects low-rank matrices (rank r typically 4–64) while freezing the original model weights, with a parameter count of only 0.1%–1% of full fine-tuning74% similarVerifiedLoRA (Low-Rank Adaptation) is a parameter-efficient fine-tuning method proposed by Microsoft Research in 2022.73% similarUnverifiedLoRA是一种轻量级微调技术，其技术源头可追溯至微软研究院2021年发表的同名论文，核心思想是用两个小矩阵的乘积近似表示神经网络权重的更新量72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/606954API
curl https://kongchang.com/api/v1/knowledge/claims/606954MCP
get_claim(id=606954)