mehilshah/ICSE26-RepGen: First release of RepGen

RepGen v0.1 Release Notes

Date: January 15, 2026 Version:: 0.1 (Initial Release)

Overview

This is the initial release of RepGen, an automated tool designed to reproduce bugs in Deep Learning (DL) systems. This release supports the methodology detailed in the associated research submission to ICSE 2026. RepGen automates the generation of reproduction scripts to facilitate fault localization and reliability testing in DL frameworks.

Key Features

• Automated Reproduction: Generates executable scripts to reproduce reported crashes and failures in DL models without manual intervention.
• Crash Analysis: Parses stack traces and issue reports to identify potential root causes in libraries such as TensorFlow and PyTorch.
• Fault Localization: Provides preliminary localization data to assist in identifying the specific modules or layers responsible for the failure.
• Environment Replication: Captures and configures the necessary software environment (dependencies, library versions) to ensure consistent reproduction results.

System Requirements

• Language: Python 3.8+
• OS: Linux (Ubuntu 20.04+ recommended) or macOS
• Dependencies: torch, tensorflow, pandas, numpy (see requirements.txt for specific versions)

Installation

git clone https://github.com/mehilshah/ICSE26-RepGen.git
cd ICSE26-RepGen
pip install -r requirements.txt
bash RepGen.sh