This repository contains a reimplementation of research work I conducted as a Graduate Research Intern in the CCS-3 Division at Los Alamos National Laboratory. This work contributed to the NAACL 2025 conference paper: "LLM-Assisted Translation of Legacy FORTRAN Code to C++: A Cross-Platform Study".

Important Note: This is a reimplementation based on my research at Los Alamos. The original code and specific data remain property of the laboratory.

Legacy Fortran code remains prevalent in scientific computing applications, but modern development often favors C++. This project evaluated the capabilities of current Large Language Models (LLMs) to assist in this translation process, potentially accelerating legacy code migration and modernization efforts.

• Designed and implemented a comprehensive framework to evaluate 12 open-source LLMs (7B-34B parameters)
• Used controlled zero-shot prompting experiments to assess code translation capabilities
• Developed standardized test cases covering various Fortran coding patterns and complexities

• Implemented CodeBLEU metrics to evaluate structural similarity between LLM-generated C++ code and ground truth translations
• Created custom evaluation metrics to assess functional equivalence and performance characteristics
• Benchmarked translation accuracy across different model sizes and architectures

• Created advanced visualizations to analyze LLM consistency and error patterns
• Identified common failure modes and logical approaches in code translation tasks
• Conducted comparative analysis between zero-shot and session-maintained prompting strategies

• Implemented an agentic feedback system with compiler error integration
• Developed methodology to improve code translation accuracy by incorporating compiler feedback
• Tested various feedback loop implementations to maximize improvement

This work contributed to the following paper: Nishath Rajiv Ranasinghe, Shawn M. Jones, Michal Kucer, Ayan Biswas, Daniel O’Malley, Alexander Most, Selma Liliane Wanna, and Ajay Sreekumar. (2025). "LLM-Assisted Translation of Legacy FORTRAN Code to C++: A Cross-Platform Study". North American Chapter of the Association for Computational Linguistics (NAACL 2025).

• Python for framework implementation and analysis
• Hugging Face Transformers for model access and inference
• Ollama models as open-source LLMs for prompting experiments
• CodeBLEU and custom metrics for evaluation
• Matplotlib/Seaborn/Plotly for advanced visualizations
• GCC and GFortran compilers for validation and feedback mechanisms

This reimplementation is made available under the MIT License.