Python-to-Rust Function Conversion Model

This project was a part of my internshit at Five Angstrom

Problem Statement

Python and Rust are fundamentally different: Python is dynamically typed and interpreted, while Rust is statically typed and compiled. This makes direct function translation between the two challenging. Key differences include typing, memory management, syntax, and error handling. The goal is to build a model that can convert Python functions into safe, efficient, and idiomatic Rust code.

What I did

• Researching about the available models out there
• Dataset creation and unit testing
• Fine-tuning Encoder-Decoder based Model
• Applying Reinforcement Learning and compared the results

Agenda

The model aims to:

• Parse Python Functions: Understand and extract the logic, structure, and types from Python code.
• Infer Types: Analyze variables and function behavior to infer static types required by Rust.
• Translate Syntax: Convert Python syntax into equivalent, idiomatic Rust code.
• Generate Rust Code: Output clean, type-safe, and compile-ready Rust functions that mirror the original Python logic.

Technologies/APIs used

• Python 3.12.x
• Transformers
• trl
• CodeBLEU (evaluation parameter)

Results

• Final Fine-tuned model gave the CodeBLEU score of 0.87229 and accurately transated 66 out of 100 functions.

Project Collaborators:

• Abraham Rebairo J: Supervisor and Guide
• Niket Girdhar (me)