Flexible Weight Tuning and Weight Fusion Strategies for Continual Named Entity Recognition (Findings of ACL 2024)

This repository contains all of our WT&WF code for ExtendNER and CFNER baselines. We sincerely thank the help of Zheng et al.'s repository.

Overview of the directory

• bert-base-cased/: the directory of configurations and PyTorch pretrained model for bert-base-cased
• config/ : the directory of configurations for our method
• datasets/ : the directory of datasets
• experiments/ : the directory of training logs from different runs
• src/ : the directory of the source code
• main_CL.py : the python file to be executed

.
├── bert-base-cased
├── config
│   ├── conll2003
│   ├── ontonotes5
│   ├── i2b2
├── datasets
│   └── NER_data
│       ├── conll2003
│       ├── i2b2
│       └── ontonotes5
├── experiments
|   └── xxx.pth
├── main_CL.py
└── src
    ├── config.py
    ├── dataloader.py
    ├── model.py
    ├── trainer.py
    ├── utils_plot.py
    └── utils.py

Step 1: Prepare your environments

Reference environment settings:

python             3.7.13
torch              1.12.1+cu116
transformers       4.14.1

Download bert-base-cased to the directory of bert-base-cased/

Download base models to the directory of experiments/

Step 2: Run main_CL.py

Specify your configurations (e.g., ./config/i2b2/fg_1_pg_1/i2b2_distill_WF_WT.yaml (ExtendNER+WT&WF)) and run the following command

CUDA_VISIBLE_DEVICES=0 nohup python3 -u main_CL.py --exp_name i2b2_1-1_distill_WT_WF --exp_id 1 --cfg config/i2b2/fg_1_pg_1/i2b2_distill_WF_WT.yaml 2>&1 &

Then, the results as well as the model checkpoint will be saved automatically in the directory ./experiments/i2b2_1-1_distill_WT_WF/1/