Group Project · 2026
A machine learning project focused on using open-source Large Language Models to translate C code into Rust. The project explored model selection, dataset preparation, fine-tuning, and evaluation using CodeT5 and Qwen2.5-Coder-7B-Instruct.
01 · Overview
The goal was to develop a model capable of translating C source code into functionally equivalent Rust code. We investigated open-source code models and fine-tuned them for this specific translation task.
We initially experimented with CodeT5 before moving to Qwen2.5-Coder-7B-Instruct. This change required us to revisit our training approach and prepare a suitable dataset for fine-tuning.
02 · My Role
I helped research potential models and worked on fine-tuning them. I also contributed to the analysis and writing using the information gathered throughout the project. We ultimately wrote a ten-page paper documenting our work and results.
03 · Team
Lead Tuner and Programmer
Researcher, Programmer, and Testing
Programming
04 · Technology
05 · Research
The following resources were used as references throughout the development of the project.
06 · Results
After encountering limitations with CodeT5, we moved to Qwen2.5-Coder-7B-Instruct and fine-tuned it using LoRA. A significant part of the process involved preparing, checking, and validating the training and validation data.
The resulting model was able to perform C-to-Rust translation and provided measurable improvements that we evaluated and documented in our final research paper.
View Written Paper