Exploring Mode Connectivity for Pre-trained Language Models

TL;DR

This paper investigates the geometric connections between different minima of pre-trained language models using mode connectivity, revealing how hyperparameters, training methods, and data influence these connections and what this implies for model adaptation.

Contribution

It provides the first empirical analysis of mode connectivity in PLMs, exploring how various factors affect the geometric relationships between minima during pre-training and adaptation.

Findings

Hyperparameters and tuning methods significantly influence mode connectivity.

Mode connectivity evolves during pre-training, affecting downstream adaptation.

Understanding these connections can improve PLM fine-tuning strategies.

Abstract

Recent years have witnessed the prevalent application of pre-trained language models (PLMs) in NLP. From the perspective of parameter space, PLMs provide generic initialization, starting from which high-performance minima could be found. Although plenty of works have studied how to effectively and efficiently adapt PLMs to high-performance minima, little is known about the connection of various minima reached under different adaptation configurations. In this paper, we investigate the geometric connections of different minima through the lens of mode connectivity, which measures whether two minima can be connected with a low-loss path. We conduct empirical analyses to investigate three questions: (1) how could hyperparameters, specific tuning methods, and training data affect PLM's mode connectivity? (2) How does mode connectivity change during pre-training? (3) How does the PLM's task knowledge change along the path connecting two minima? In general, exploring the mode connectivity of PLMs conduces to understanding the geometric connection of different minima, which may help us fathom the inner workings of PLM downstream adaptation.