My-GPIoT-setup

GPIoT Modified Setup for Local Reproduction & Enhancement

This repository contains my modifications, and documentation for reproducing and extending the GPIoT project.

📌 Original Project

Title: GPIoT: Tailoring Small Language Models for IoT Program Synthesis and Development
Authors: Leming Shen, Qiang Yang, Xinyu Huang, Zijing Ma, Yuanqing Zheng
Conference: ACM SenSys 2025
License: MIT
Original Repo: https://github.com/lemingshen/GPIoT
Paper: https://lemingshen.github.io/assets/publication/conference/GPIoT/paper.pdf
Project: GPIoT: Tailoring Small Language Models for IoT Program Synthesis and Development

⚠️ This is NOT the official repository. I do not distribute the original model weights or datasets. Please download them from the authors' provided links.

🧩 My Contributions

• ✅ Provided a detailed dependency setup tailored to my hardware environment

• ✅ Modified task_generation.py to optimize task generation logic

• ✅ Modified code_generation.py to improve code generation workflow

• ✅ Modified fine_tune.py to enhance training stability and adaptability

  These modifications improve local reproducibility, training stability, and code generation effectiveness.

📥 Installation

⚙️ My Setup Instructions

Hardware Environment

• GPU: NVIDIA RTX 3090
• Environment: Conda virtual environment

Key Packages

The project was tested under the following major dependencies:

Package	Version
torch	2.1.0+cu118
torchvision	0.16.0+cu118
torchaudio	2.1.0+cu118
transformers	4.36.0
accelerate	0.25.0
datasets	2.15.0
peft	0.12.0
trl	0.7.6
bitsandbytes	0.41.3
scipy	1.15.3
pandas	2.3.2
matplotlib	3.7.5
wandb	0.21.3
sentencepiece	0.1.99

Installation

# 1. Clone the repository
git clone https://github.com/yourname/your-gpiot-modified
cd your-gpiot-modified

# 2. Create and activate conda environment
conda create -n gpiot-env python=3.10 -y
conda activate gpiot-env

# 3. Install dependencies
pip install -r requirements.txt