Source codes for the baseline models of Call for Customized Conversation: Customized Conversation Grounding Persona and Knowledge, accepted at AAAI-22.

We trained the models under the setting of python==3.7 and torch==1.5.0, with one RTX8000 GPU. Also, our codes are built on the codes of huggingface, and we utilized pytorch-ignite from pytorch in ignite folder.

1.Make a virtual environment

$conda create -n ENV_NAME python=3.7

2.Install pytorch==1.5.0

$conda install pytorch==1.5.0 torchvision==0.6.0 cudatoolkit=10.2 -c pytorch

3.Install the required libraries.

$pip install -r requirements.txt

This data is the modified version of the original data (which is reported in the paper) after ethical inspection.

You should create directories named infer_log_focus, train_log_focus, test_log_focus, models, data under FoCus folder.

We put train, valid, test files of the dataset in the data folder. (The test set will be available after March 2022.)

The project directory should follow this directory structure:

📦FoCus
┣ 📂data
┃ ┗ 📜train.json
┃ ┗ 📜valid.json
┣ 📂ignite
┣ 📂infer_log_focus
┣ 📂models
┣ 📂python_tf_idf
┣ 📂test_log_focus
┣ 📂train_log_focus
┣ 📜classification_modules.py
┣ 📜data_utils.py
┣ 📜evaluate_test.py
┣ 📜evaluate_test_ppl.py
┣ 📜inference.sh
┣ 📜inference_test.py
┣ 📜LICENSE
┣ 📜README.md
┣ 📜requirements.txt
┣ 📜test.sh
┣ 📜train.sh
┣ 📜train_focus.py
┗ 📜utils_focus

Uncomment the command lines in the train.sh file, to start training the model.

$ sh train.sh 

Uncomment the command lines in the test.sh file, to evaluate the model on the test set.

$ sh test.sh

Uncomment the command lines in the inference.sh file, to generate utterances with the trained models.

$ sh inference.sh

We are going to hold the 1st workshop on Customized Chat Grounding Persona and Knowledge in Octorber 2022. Stay tuned for our latest updates!

Written by Yoonna Jang.

(c) 2021 NCSOFT Corporation & Korea University. All rights reserved.