A parameter-efficient adaptation method that trains low-rank updates while keeping base model weights frozen.
Sign in to saveLow-Rank Adaptation (LoRA) adapts a pretrained model by learning small low-rank matrices instead of updating all original weights. The method reduces the trainable parameter count and associated training memory. A LoRA adapter represents a task-specific update; it is not necessarily a standalone model.
Research paper first submitted to arXiv on 2021-06-17.
Useful for comparing fine-tuning techniques and downloadable adapters. Compatibility depends on the base architecture and the weights the adapter was trained against. Catalog note: this is a source-verified reference entry, not a measured current-trend signal. No search-volume or growth claim is made.
Pro opens the same-day prioritize / watch / skip table, including names that are still too new to publish here.
This name has not appeared in an imported daily report yet.