QaDialMoE

Code for paper "QaDialMoE: Question-answering Dialogue based Fact Verification with Mixture of Experts".

Dependency

• python 3.5+
• pytorch 1.0+

Usage

Download RoBERTa-Large model from https://huggingface.co/roberta-large. Place all the files into a new folder "roberta_large" under the root.

Dataset

Please see the details in data folder.

Training

To train the QaDialMoE:

python run_healthmoe.py --do_train --do_eval
python run_colloquial_t5.py --do_train --do_eval
python run_faviqmoe_dpr.py --do_train --do_eval

Acknowledgement

Our implementation is built on the source code from SaMoE. Thanks for their contributions.