Multitask Multilingual Model Adaptation with Featurized Low-Rank Mixtures

arXiv:2402.17934v1 Announce Type: new
Abstract: Adapting pretrained large language models (LLMs) to various downstream tasks in tens or hundreds of human languages is computationally expensive. Parameter-efficient fine-tuning (PEFT) significantly reduces the adaptation cost, by tuning only a small amount of parameters. However, directly applying PEFT methods such as LoRA (Hu et al., 2022) on diverse dataset mixtures could lead to suboptimal performance due to limited parameter capacity and negative interference among different datasets. In this work, we propose Featurized Low-rank …

abstract arxiv cost cs.ai cs.cl fine-tuning human language language models languages large language large language models llms lora low model adaptation multilingual parameters peft small tasks type