Unverified50% confidenceFactExact time
LoRA中A使用随机高斯初始化，B初始化为零，以确保训练开始时叠加量为零、不破坏原模型能力，推理时不引入额外延迟
1
Sources
50%
Confidence
Long-term
Relevance
7/24/2026
First Seen
Sources
零基础入门大模型微调：原理、场景与实战路径
Related Claims
VerifiedAfter LoRA training is complete, the low-rank matrices can be merged back into the original weights, introducing no additional latency during inference.75% similarVerifiedLoRA (Low-Rank Adaptation) was proposed by Microsoft in 2021, with the core idea of freezing the weights of the pretrained model and only training injected low-rank decomposition matrices.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/610905API
curl https://kongchang.com/api/v1/knowledge/claims/610905MCP
get_claim(id=610905)