🚨 News (2025-03-5): Our system paper CodeTransEngine: Ready-to-Use Backend for LLM-Based Code Translation has been accepted to the ICLR 2025 Third Workshop on Deep Learning for Code
🚨 News (2025-01-21): We're excited to share that our paper InterTrans: Leveraging Transitive Intermediate Translations to Enhance LLM-based Code Translation has been accepted to the Main Research Track of the 47th IEEE/ACM International Conference on Software Engineering (ICSE 2025)
Welcome to the documentation for CodeTransEngine. This is a ready-to-use backend for Large Language Model (LLM) based code translation across programming languages. This tool enables practitioners to translate source code across programming languages at scale, by leveraging off-the-shelf Large Language Models (LLM). This backend integrates the Tree of Code Translation (ToCT) algorithm used in the InterTrans Paper can be used with few-shot prompting, agents or newer algorithms.
CodeTransEngine serves as a backend for code translation, helping you save time and effort in building such infrastructure from scratch. It is extensible and high-performant due to its concurrent architecture and other optimizations.
- 🧠 Multiple algorithms (InterTrans, Direct Translation, Few-shot Prompting and more)
- ⚡ Efficient inference using vLLM as backend and OpenAI Compatible APIs
- 🌐 Distributed inference supported
- 🛡️ Safe and containerized code execution
- 📊 Automatic translation evaluation using test-cases
- 🔧 Extensible to new datasets, prompts and translation algorithms
- ♻️ Configurable cache for resource saving
- 🚆 Fully concurrent architecture for maximum throughput or sequential for resource saving
- 🔗 Can be used standalone or integrated into existing workflows for code translation
Please see the Documentation Page
CodeTransEngine is designed to support a variety of user cases, ensuring seamless integration into a wide range of code translation workflows. This adaptability allows it to function both as a standalone tool and as a modular component within larger pipelines for automated code translation. For instance, it can serve as the backend for whole-repository translation initiatives, where tasks such as file listing, planning, and program slicing can be executed outside of the engine, or it can be used standalone to translate source code
If you use this tool for academic purposes, please considering citing our system paper:
@inproceedings{macedo2025codetransengine,
title={CodeTransEngine: Ready-to-use Backend for LLM-based Code Translation},
author={Macedo, Marcos and Tian, Yuan and Adams, Bram},
booktitle={ICLR 2025 Third Workshop on Deep Learning for Code}
}The contents of this repository are licensed under the MIT License. For detailed information, please refer to the LICENSE.md file.