DiffQRe

The code is for our ACL2022 paper: On the Robustness of Question Rewriting Systems to Questions of Varying Hardness

1. Environment

a. conda create --name qre python=3.7

b. pip install pytorch_lightning==0.8.0 or pip install pytorch_lightning==1.0.4

c. pip install transformers==3.3.1

2. to Obtain Adapter S on BART

sh ./bin/finetune_shared.sh

3. to Train Private Models

sh ./bin/finetune_shared.hard_0.sh

sh ./bin/finetune_shared.hard_1.sh

sh ./bin/finetune_shared.hard_2.sh

4. to Train SLAF

sh ./bin/finetune.load_adapter.domain_weight_soft.sh

5. to Train SLAD

sh ./bin/finetune.load_adapter.distill.sh

6. to Do Evaluation

sh ./bin/eval_bleu.sh

7. Data Downloading

Cannard: https://sites.google.com/view/qanta/projects/canard

QReCC: https://github.com/apple/ml-qrecc