CMR-Sum

This is the source code for "Cross-Modal Retrieval-Enhanced Code Summarization based on Joint Learning for Retrieval and Generation"

Package Requirement

To run this code, some packages are needed as following:

torch==2.0.0
torchvision==0.15.0
numpy==1.23.5
tqdm==4.65.0
transformers==4.27.1
nltk==3.8.1
faiss==1.7.2

Moreover, a jdk is also needed to run the evaluation script for METEOR.

Dataset

Prepare dataset for java and python:

cd dataset
unzip data.zip

Training and testing

For the Java dataset:

bash run_java.sh

For Python data:

bash run_python.sh

The model checkpoints, generated results, retrieved results and reference will be saved in the "saved_models" folder.