RFM

The code for the paper: "Feedback is important: response-aware feedback mechanism for background based conversation."

Requirements

• python 3.7
• pytorch 1.7.0

Datasets

• Download the raw data version of Holl-E, and put the raw data files (train_data.json, dev_data.json and test_data.json）in the directory /dataset/raw_data.
• Then, run the preprocessing script:

python Prepare_holl.py

• Download the glove.6B.300d.txt and put it in /dataset/oracle and /dataset/mixed.

Run training, validation, and testing

To train or test your model, run:

python -m torch.distributed.launch --nproc_per_node=num_GPU Run_RFM.py --mode='train/test'

Addition

More descriptions will be released in a few days...