This directory contains configuration files for FusionBench. These configurations are essential for setting up and managing various algorithms and their hyperparameters.
FusionBench employs a modular configuration system, which is divided into three primary groups:
- Method Configuration: Defines the fusion algorithm and its associated hyperparameters.
- Model Pool Configuration: Manages the models involved in the fusion process, including datasets, tokenizers, preprocessors, and other related resources.
- Task Pool Configuration: Specifies the tasks and their corresponding datasets used for evaluating the fused models.