TransferTOD

The model repository of paper "TransferTOD: A Generalizable Chinese Multi-Domain Task-Oriented Dialogue System with Transfer Capabilities"

All the code and data can be found in https://github.com/KongLongGeFDU/TransferTOD

Paper: https://aclanthology.org/2024.emnlp-main.710.pdf

Data Info

Overall statistics of TransferTOD dataset are as follows:

	Train	ID Test	OOD Test
#Domain	27	27	3
#Slot	188	188	27
#Dialogue	4320	540	600
#Turns	28680	3585	3700
#Slots/Dialogue	10.3	10.3	9.7
#Tokens/Turn	66.4	66.4	76.8

ID Test means In-Domain test and OOD Test means Out-of-Domain test. The domains of the test set are Water-Delivery, Sanitation, and Courier.

Guide

All the data used in two-staged finetuning and the raw data of TransferTOD is included in directory ./data. For each version, train.json is a mixed data of train_slot.json and equivalent amounts of ./data/raw_data/belle_data/belle_filtered_950k_train.jsonl

For full fine-tuning, run ./fine_tune/scripts/finetune_full.sh, while for lora fine-tuning, run ./fine_tune/scripts/finetune_lora.sh.

For inference and evaluation with the TransferTOD test set, run ./inference/inference_and_eval.sh.

Citation

If you find this project useful in your research, please cite:

@inproceedings{DBLP:conf/emnlp/ZhangHWLZDSDZYZ24,
  author       = {Ming Zhang and
                  Caishuang Huang and
                  Yilong Wu and
                  Shichun Liu and
                  Huiyuan Zheng and
                  Yurui Dong and
                  Yujiong Shen and
                  Shihan Dou and
                  Jun Zhao and
                  Junjie Ye and
                  Qi Zhang and
                  Tao Gui and
                  Xuanjing Huang},
  editor       = {Yaser Al{-}Onaizan and
                  Mohit Bansal and
                  Yun{-}Nung Chen},
  title        = {TransferTOD: {A} Generalizable Chinese Multi-Domain Task-Oriented
                  Dialogue System with Transfer Capabilities},
  booktitle    = {Proceedings of the 2024 Conference on Empirical Methods in Natural
                  Language Processing, {EMNLP} 2024, Miami, FL, USA, November 12-16,
                  2024},
  pages        = {12750--12771},
  publisher    = {Association for Computational Linguistics},
  year         = {2024},
  url          = {https://aclanthology.org/2024.emnlp-main.710},
  timestamp    = {Thu, 14 Nov 2024 17:20:55 +0100},
  biburl       = {https://dblp.org/rec/conf/emnlp/ZhangHWLZDSDZYZ24.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}

TransferTOD-7B

Model card