README.md

metadata

task_categories:
  - graph-ml
  - text-generation
tags:
  - code
size_categories:
  - 10K<n<100K

Here is a 100 repositories from Codearena parsed as graph of functions with different types of relationships: Parent Contracts, High-Level Calls, Internal Calls, Library Calls, Low-Level Calls.

Dataset can be used for tasks like code clone detection (similar bugs search), code reconstruction with context pruning. Dataset is ready to be used in GNN.

You can see usage of dataset as RAG at github

Structure

auditits_info.json <-main file for easy paths navigantion

/processed repositories

/repository1

    /file1
    
        /function1(node) <- contains code, relationships(edges) with other functions
        
        /function2
        
    /file2
    
        /function3
        
        /function4
        
/repository2

    ...
    

/reports

/repository1

    /vulnerability1-code.txt <-actual code with bug
    
    /vulneralbility1-explain.txt <-human explanation

    ...
    
...