v0.1.0

Public release of LoTR, a novel approach for parameter-efficient fine-tuning of LLMs which represents a gradient update to parameters in a form of tensor decomposition.

arXiv:2402.01376