low-rank adaptation

A technique for fine-tuning large models by introducing low-rank matrices into existing layers, which reduces the number of parameters to be updated while maintaining performance.

39 papers