LoRA Low-Rank_Adaptation_of_Large_Language_Models LoRA_for_Convolution_Network LECO
A method of finetuning adapted from a method previously used for language models. This method has multiple advantages, including faster training and smaller filesizes. Many training scripts now support this method out of the box.
Paper: https://arxiv.org/abs/2106.09685
Adaption for diffusion models: https://github.com/cloneofsimo/lora
Locon/LyCORIS expands upon this method, using further training enhancements and/or training more layers of the diffusion model (LoRA, in contrast, trains fewer diffusion model layers, something observed to be viable in the original LoRA paper). https://github.com/KohakuBlueleaf/LyCORIS
Posts with this metatag indicate they utilize at least one LoRA-trained model (or Locon/LyCORIS/etc).
The following tags are aliased to this tag: locon (learn more).
