read-and-select

Environment

conda activate -n res python=3.9
conda activate res
pip install -r requirements.txt

Data and Checkpoints

Please see the README.md files in different folders to download the corresponding data and checkpoints.

Evaluate with Our Checkpoints

After downloading the data and checkpoints, you can use the command below to replicate our results reported in the paper.

python run_disambiguation_attention.py --do_eval \
--model model_disambiguation/zeshel_disambiguation_attention.pt \
--transformer_model roberta-base \

Train Res

You can train your own model with the command below.

python run_disambiguation_attention.py --do_eval \
--do_train \
--model model_disambiguation/zeshel_disambiguation_attention.pt \
--transformer_model roberta-base \
--cand_num 10
--batch 4
--gpus 0,1,2,3

🚨: if you follow our experiment setting to set cand_num=56, it takes 4*40G A-100.