DARE the Extreme: Revisiting Delta-Parameter Pruning For Fine-Tuned Models

TL;DR

This paper enhances delta-parameter pruning for fine-tuned models by introducing DAREx, which improves performance at high pruning rates and combines with existing fine-tuning techniques, reducing redundancy and response times.

Contribution

The paper proposes DAREx, an improved delta-parameter pruning method with algorithmic enhancements and integration with in-training regularization, addressing limitations of previous DARE approaches.

Findings

DAREx-q significantly improves high-rate pruning performance.

Combining DAREx with in-training regularization enhances pruning effectiveness.

Importance-based pruning outperforms random methods for large delta parameters.

Abstract

Storing open-source fine-tuned models separately introduces redundancy and increases response times in applications utilizing multiple models. Delta-parameter pruning (DPP), particularly the random drop and rescale (DARE) method proposed by Yu et al., addresses this by pruning the majority of delta parameters--the differences between fine-tuned and pre-trained model weights--while typically maintaining minimal performance loss. However, DARE fails when either the pruning rate or the magnitude of the delta parameters is large. We highlight two key reasons for this failure: (1) an excessively large rescaling factor as pruning rates increase, and (2) high mean and variance in the delta parameters. To push DARE's limits, we introduce DAREx (DARE the eXtreme), which features two algorithmic improvements: (1) DAREx-q, a rescaling factor modification that significantly boosts performance at high pruning rates (e.g., >30 % on COLA and SST2 for encoder models, with even greater gains in decoder models), and (2) DAREx-L2, which combines DARE with AdamR, an in-training method that applies appropriate delta regularization before DPP. We also demonstrate that DAREx-q can be seamlessly combined with vanilla parameter-efficient fine-tuning techniques like LoRA and can facilitate structural DPP. Additionally, we revisit the application of importance-based pruning techniques within DPP, demonstrating that they outperform random-based methods when delta parameters are large. Through this comprehensive study, we develop a pipeline for selecting the most appropriate DPP method under various practical scenarios.