ACToR

🎯 Overview

📁 Project Structure

ACToR/
├── README.md                # Project documentation
├── classifiers/             # Critical Token classifiers 
├── weights/                 # Position-aware weights
├── datasets/                # Benchmarks and training datasets
├── repositories/            # Code repositories for training and testing
└── src/                     # Source code directory
    ├── pipeline.py
    ├── critoken.py
    ├── train.py
    ├── data/                # Data processing pipeline
    │   ├── __init__.py
    │   ├── repo.py
    │   ├── task.py
    │   └── process/         # Data processing modules
    │       ├── __init__.py
    │       ├── data.py
    │       ├── window.py
    │       ├── vector.py
    │       ├── search.py
    │       ├── prompt.py
    │       └── utils.py
    └── server/              # Model server integration
        ├── __init__.py
        ├── classifier.py
        └── llm.py

🛠️ Quick Start

1. Prerequisites & Installation

pip install -r requirements.txt

2. Repository Setup

repositories/
├── codereval/
│   └── python/              # CoderEval Python repositories
├── repoexec/
│   └── test-app/            # RepoExec test repositories
└── repost/
    └── train/               # RepoST training repositories

⚡ Usage

python src/pipeline.py repo --benchmark 'repoexec' # alternative 'codereval-python', 'repost_train'

python src/pipeline.py train --task_type 'data' --model_name 'model_name' # e.g. codellama-7b-hf

python src/pipeline.py train --task_type 'classifier' --model_name 'model_name' # e.g. codellama-7b-hf

python src/pipeline.py task token --model_name 'model_name' --benchmark 'repoexec'