Question-NLI

This is the course project of F033569 NLU of SJTU.

Requirements

• Python3.*
• Pytorch 0.4.1 (not test in higher version)
• pytorch_pretrained_bert
• tqdm (to monitor the training progress)
• GPU (indispensable, since the model is saved with CUDA format)

To train

Put your data into the /data folder and the pretrained model in /data/.cache folder, then simply run

sh train.sh

or

python3 main.py --mode train --lower_case --gpu --gpu_id 0 --max_seq_len 128 --num_epochs 20 --lr 2e-5 --batch_size 32 --warmup_proportion 0.1

To test

Set the paths of your models in model_list.txt (note that the file should be end with a new line), then run

sh test.sh