What Does This Mean?
Low-Rank Adaptation, a method that adapts a large pretrained model by freezing its original weights and training small added low-rank matrices, which cuts the number of trainable parameters sharply. In hobby AI communities a "LoRA" also means the small add-on file produced this way.
REAL-WORLD EXAMPLE
""She trained a small LoRA on her drawings and applied it to the base model.""
LORE & ORIGIN
The 2021 paper "LoRA: Low-Rank Adaptation of Large Language Models" by Edward J. Hu and colleagues was submitted to arXiv on June 17, 2021; it adds trainable low-rank matrices to each Transformer layer while the pretrained weights stay fixed. Wikipedia's fine-tuning article describes LoRA as an adapter-based technique that can come close to full fine-tuning results with far fewer trainable parameters. The community meaning of "a LoRA" as a shareable add-on file was not documented in the pages opened.
MENTIONS OVER TIME
Radar observation data not yet available for this term.
RELATED SEARCHES
TOP PLATFORMS
Not enough Radar data yet to show a platform breakdown.
WHEN DID YOU FIRST HEAR THIS?
CLICK AN OPTION BELOW TO CAST YOUR VOTE.
[ VOTE TO REVEAL COMMUNITY RESULTS ]