Fine-tuning

Standard Fine-tuning

Thanks a lot for OpenNRE.

Preparation

All datasets need to be placed in the benchmark folder.

Running

>> python train_supervised_bert.py --dataset=SciERC/10-1 --pretrain_path=dmis-lab/biobert-large-cased-v1.1

• --dataset: the directory of the dataset to be trained
• --pretrain_path: PLM, defaulting to roberta-large

Balancing

• Re-sampling datasets refers to README.
• Re-weighting Loss: --use_loss

Data Augmentation

• DA refers to README.

Self-training

• Assign pseudo labels: --labeling True
• Combine pseudo-labeled data and gold-labeled data by using self-train_combine.py
• Train the student model: --stutrain True