LISA: Layerwise Importance Sampling for Memory-Efficient Large Language Model Fine-Tuning

arXiv:2403.17919v1 Announce Type: new
Abstract: The machine learning community has witnessed impressive advancements since the first appearance of large language models (LLMs), yet their huge memory consumption has become a major roadblock to large-scale training. Parameter Efficient Fine-Tuning techniques such as Low-Rank Adaptation (LoRA) have been proposed to alleviate this problem, but their performance still fails to match full parameter training in most large-scale fine-tuning settings. Attempting to complement this deficiency, we investigate layerwise properties of LoRA on fine-tuning tasks …

abstract arxiv become community consumption cs.ai cs.cl cs.lg fine-tuning importance language language model language models large language large language model large language models llms lora low low-rank adaptation machine machine learning major math.oc memory memory consumption model fine-tuning sampling scale training type