CIDER

Resources for paper "Inconsistent dialogue responses and how to recover from them".

Dataset

The proposed dataset CIDER is located in folder data/, where there are two files according to the data source. The English version will come soon.

Model Training and Testing

Script train.py is for training and test_datasets.py for testing. Task to conduct could be chosen by parameter task, here is an example for resolve_turn task.

Multi-GPU Training

num_gpu=4
lr=3e-4
task=resolve_turn
arch=t5
model_sig=uer/t5-base-chinese-cluecorpussmall
python -m torch.distributed.launch --use_env --nnodes=1 --nproc_per_node=${num_gpu} train.py \
    --debug=n \
    --dataset=both \
    --task=${task} \
    --add_tcon=y \
    --add_cdconv=n \
    --add_stance=n \
    --add_ocnli=n \
    --arch=${arch} \
    --save_dir=exp/resolver/${task}_${arch}_${lr} \
    --ckpt_dir='' \
    --model_sig=${model_sig} \
    --learning_rate=${lr} \
    --batch_size=16 \
    --patience_num=3 \
    --validate_on_tcon=y \
    --with_explanation=n

Testing

task=resolve_turn
arch=t5
lr=3e-4
python test_datasets.py \
    --debug=n \
    --task=${task} \
    --arch=${arch} \
    --save_dir=exp/resolver/${task}_${arch}_${lr} \
    --model_dir=exp/resolver/${task}_${arch}_${lr} \
    --with_explanation=n

Citation

Please cite our paper if you find the codes or dataset helpful using this BibTeX:

@article{zhang2024inconsistent,
  title={Inconsistent dialogue responses and how to recover from them},
  author={Zhang, Mian and Jin, Lifeng and Song, Linfeng and Mi, Haitao and Yu, Dong},
  journal={arXiv preprint arXiv:2401.10353},
  year={2024}
}