The Struggles of LLMs in Cross-lingual Code Clone Detection

Description:

This project explores whether large language models (LLMs), baselines and traditional classifiction algorithms can identify code clones across different programming languages.

This work aims to:

• Evaluate the effectiveness of LLMs in cross-lingual code clone detection.
• Compare LLM performance to baselines and traditional methods.
• Uncover the best approach for identifying similar code across languages.

Content of the repository

This repository contains the following files:

• data_selection: This directory contains all the code to select the subsets
• data: This directory contains the subset of XLCoST and CodeNet used in the experiments
• classifier: This directory contains all the code and data for the classification part
• results: This directory contains all the results for each dataset and each LLM
• get_embeddings.py : permits to generate the vectors for each code snippet
• sp, se, sct : permit to run experiment with the gpt-3.5-turbo depending on the experiment
• llama2_inf, falcon_inf, starchat_inf, starcoder_inf : permit to run experiment with the llama-2-7b-chat-hf, falcon-7b-instruct, starchat-beta and starcoder2-15b-instruct