This repo contains the code and data for HyperNetwork-based Decoupling to Improve Model Generalization for Few-Shot Relation Extraction (EMNLP-2023).
- Python 3.7.4
- PyTorch 1.7.0
- CUDA 10.2
The expected structure of files is:
|-- checkpoint
|-- data
|-- fewshot_re_kit
|-- logs
|-- models
|-- train_demo.py
|-- run_bert.sh
Our model HND is placed in the HND.py file.
Our two-stage training strategy can be found on lines 220 to 320 of the framework.py file in the fewshot_re_kit folder.
You can train a N-way-K-shot model by:
sh run_bert.sh
In the file run_bert.sh, we can modify the settings and hyper-parameters of the model.