Unverified50% confidenceFactExact time
During LoRA inference, W' = W₀ + (α/r)BA can be pre-computed to merge the LoRA branch into the original weights, achieving zero additional inference latency.
1
Sources
50%
Confidence
Long-term
Relevance
9/10/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedLoRA推理时可将BA与原权重合并，不增加任何推理延迟，通常只应用于注意力层的Q、V权重矩阵73% similarUnverifiedLoRA 由 Hu 等人在 2021 年提出，通过为每层注入可训练的低秩分解矩阵实现参数更新 ΔW = BA，将可训练参数量压缩至原始模型的 0.1%-1%68% similarUnverifiedLoRA attaches two low-rank matrices A and B in parallel alongside each target weight matrix, where A is randomly initialized and B is initialized to zero; only these two small matrices are trained, and during inference the product A×B is added back to the original weights.67% similarUnverifiedDuring LoRA training, the pre-trained model weights W₀ are frozen and only the two matrices B and A are trained, with the forward propagation formula H = W₀X + (α/r)BAX.67% similarVerifiedLoRA achieves fine-tuning by inserting low-rank decomposition matrices alongside the model's weight matrices, typically requiring training of only 0.1%–1% of the original parameter count.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/888470API
curl https://kongchang.com/api/v1/knowledge/claims/888470MCP
get_claim(id=888470)