Carbon-Taxed Transformers Replication Package

This repository provides a comprehensive replication package for the study titled Carbon-Taxed Transformers: A Green Compression Pipeline for Overgrown Language Models. It includes neural architecture search (NAS) scripts, training/evaluation pipelines, and empirical analysis figures for three tasks: code clone detection, code generation, and code summarization.

Repository Structure

The repository is organized into the following directories:

1. clone_detection/

This folder contains code and data for conducting NAS and training models for the code clone detection task.

• code_clone_nas.py: Entry point to run NAS. Execute via:

  python code_clone_nas.py
  

• Subdirectories:

  • compress/: Contains scripts for training and evaluating student models. Execution instructions are provided within the folder.
  • data/: Contains all data related to code clone detection.
  • finetune/: Contains scripts for training and evaluating teacher models. Instructions are included within the folder.

2. code_generation/

This folder contains code and training pipelines for the code generation task.

• code_gen_nas.py: Entry point to run NAS for code generation. Execute via:

  python code_gen_nas.py
  

• Subdirectories:

  • compress/: Scripts for student model training and evaluation.
  • finetune/: Scripts for teacher model training and evaluation.

Execution instructions are available inside the respective folders.

3. code_summarization/

This folder provides code and utilities for the code summarization task.

• code_sum_nas.py: Entry point to run NAS for code summarization. Execute via:

  python code_sum_nas.py
  

• Subdirectories:

  • compress/: Scripts for student model training and evaluation.
  • data/: All task-related data.
  • finetune/: Scripts for teacher model training and evaluation.

Detailed usage instructions are provided within the respective subfolders.

5. requirements.txt

This file lists the Python package dependencies required to run the code in this repository. Install them using:

pip install -r requirements.txt