Online-CL-LLMs

This repository contains the official implementation for our ICML 2025 paper: Exploiting Presentative Feature Distributions for Parameter-Efficient Continual Learning of Large Language Models.

Training

Generate the training script by executing:

python gen_script_new_{benchmark}_{model}.py

Then run the resulting script to start the training.

Evaluation

Compute key metrics including: Average Performance (AP), Forgetting Rate (F.Ra), Forward Transfer (FWT) and Backward Transfer (BWT). Execute the following command:

python score.py your_result_path single_result_path 

Credits

The code of this repository partly relies on SAPT and we would like to show our sincere gratitude to authors of it.