sogang-isds / TOATOD

Task-Optimized Adapters for an End-to-End Dialogue System Paper Code

Task-Optimized Adatper for an End-to-End Task Oriented Dialog

This repository contains the code for the paper Task-Optimized Adapter for an End-to-End Task Oriented Dialog (Findings of ACL 2023).

Prerequisite

Install Requirements

pip install -r requirements.txt

wandb setup

pip install wandb
wandb login
wandb init

Download & Preprocess Data

MultiWOZ 2.1

cd data/multiwoz21
bash data_preparation.sh

MultiWOZ 2.2

• MultiWOZ 2.2 Data Preprocessing First, You have to download the dataset from MultiWOZ 2.2. And then, you have to convert the dataset to the format of MultiWOZ 2.1.

cd data/multiwoz22
bash data_preparation.sh

NLU (IC) task

cd data/banking77
bash banking77_preparation.sh

cd ../clinc150
bash clinc150_preparation.sh

cd ../hwu64
bash ../hwu64_preparation.sh

Download Pre-trained Weights

cd checkpoints

wget https://pptod.s3.amazonaws.com/Pretrain/small.zip
unzip small.zip
rm small.zip

wget https://pptod.s3.amazonaws.com/Pretrain/base.zip
unzip base.zip
rm base.zip

Train & Eval of E2E TOD

small

cd E2E_TOD
bash small_run_21.sh
bash small_run_22.sh

base

cd E2E_TOD
bash base_run_21.sh
bash base_run_22.sh

Train & Eval of NLU (IC)

cd IC
bash run.sh

TOATOD Checkpoints

If you want to test our best models, download the checkpoint files from the following links.

Task	Dataset	Model (small)	Model (base)
E2E TOD	MultiWOZ 2.1	Link	Link
	MultiWOZ 2.2	Link	Link
NLU	banking77	Link	Link
	clinc150	Link	Link
	hwu64	Link	Link

Reference

We heavily referenced the code from PPTOD and would like to express our gratitude to the Amazon crew.

@article{su2021multitask,
   author = {Yixuan Su and
             Lei Shu and
             Elman Mansimov and
             Arshit Gupta and
             Deng Cai and
             Yi{-}An Lai and
             Yi Zhang},
   title     = {Multi-Task Pre-Training for Plug-and-Play Task-Oriented Dialogue System},
   booktitle = "Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (ACL)",

   year      = {2022},
   url       = {https://arxiv.org/abs/2109.14739}
}