README.md

Paragraph encoder

• If you just want to use the paragraph representation used in the paper, please download the pretrained vectors. Refer to the data section of this README for more details.

Please run the following commands from the top-level directory.

Training

python paragraph_encoder/train_para_encoder.py --data_dir data/ --src quasart|searchqa|triviaqa --embed_dir data/embeddings --model_dir model_save_dir 

To save the vectors using the pretrained models

python paragraph_encoder/train_para_encoder.py --eval_only 1 --pretrained /path/to/model_save_dir/model.mdl --src quasart|searchqa|triviaqa --save_dir data/