The valid dataset is a sub dataset of the trainning dataset, thus the model implemented by @taorui has a 0.97 accuarcy.
- download
Chinese_wmm
from GitHub - extract the files and paste them to
opennre/pretrain/chinese_wwm
- rename
bert_config.json
toconfig.json
python example/train_supervised_chinese_wwm.py \
--pretrain_path {os.getcwd()}/OpenNRE/opennre/pretrain/chinese_wwm \
--dataset people-relation
The ckpt
file will save in OpenNRE/ckpt
folder.