Download the data

Download the data from https://drive.google.com/drive/folders/1Ci-KLHKk-yP-y5fWX4_cU8bA2fL_q76e?usp=sharing

You should construct the files as follows:

.
└── data
    ├── train_2.json
    ├── dev_2.json
    └── test_2.json

Download a competition model

Download the best model of the competition from https://bounty-package.s3.cn-northwest-1.amazonaws.com.cn/biendata/226371/haihua-gpu_ver5.tar.gz

Evaluate the model on the test set

docker load < haihua-gpu_ver5.tar.gz
docker run --runtime=nvidia -v `pwd`/data:/app/data haihua_gpu_image_ver_5 -input data/test_2.json -output data/out.csv

The result will be saved in data/out.csv

you can access the model dicretly by

docker run -it --rm --entrypoint="/bin/bash" haihua_gpu_image_ver_5