Low-Rank Adaptation
A low-rank update method for adapting model weightsA method that keeps pretrained weights frozen and learns low-rank update matrices for task or domain adaptation.
Official sourceLoRA
Pronounced “low-rah”. Check its meaning and usage context with sources.
LoRA
low-rahOfficial source
The answer in 30 seconds
LoRA stands for Low-Rank Adaptation. It freezes the original model weights and trains low-rank update matrices, reducing the number of parameters that need to be optimized for an adaptation.
A parameter-efficient method for adapting a pretrained model with low-rank updates.
LoRAWhat does it stand for?
Meanings by domain
Domains and contexts
Usage contexts in the published data
Review notes
Points that prevent overclaiming
LoRA is one PEFT method. Rank, target layers, training data, and the base model affect the resulting adapter.
LoRAFrequently asked questions
Expansions and usage
- What does LoRA stand for?
- It can stand for Low-Rank Adaptation (A low-rank update method for adapting model weights). The intended meaning depends on the surrounding context.
Related terms
Explore by domain and context
Sources and verification
Verification date and sources
Status
Official source
Last verifiedAug 21, 2026
This entry is based on researched sources and item-level verification notes.
Understand the term,
then check
the context.
Expansions, meanings, domains, and evidence are shown separately so an abbreviation can be read in context.