https://github.com/wyatt-avilla/sunbird

neural network based deompiler
https://github.com/wyatt-avilla/sunbird

decompilation machine-translation neural-network python pytorch tree-sitter x86-64

Last synced: 1 day ago
JSON representation

neural network based deompiler

• Host: GitHub
• URL: https://github.com/wyatt-avilla/sunbird
• Owner: wyatt-avilla
• Created: 2024-09-04T01:07:04.000Z (23 days ago)
• Default Branch: main
• Last Pushed: 2024-09-19T18:16:19.000Z (7 days ago)
• Last Synced: 2024-09-20T14:05:18.516Z (6 days ago)
• Topics: decompilation, machine-translation, neural-network, python, pytorch, tree-sitter, x86-64
• Language: Jupyter Notebook
• Homepage:
• Size: 49.8 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: readme.md

Awesome Lists containing this project

README

        
# sunbird 🐦‍🔥

[![Ruff](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/ruff/main/assets/badge/v2.json)](https://github.com/astral-sh/ruff)

[![Checked with mypy](http://www.mypy-lang.org/static/mypy_badge.svg)](http://mypy-lang.org/)

[![python](https://img.shields.io/badge/Python-3.12-3776AB.svg?style=flat&logo=python&logoColor=white)](https://www.python.org)

[![pytorch](https://img.shields.io/badge/PyTorch-2.4.1-EE4C2C.svg?style=flat&logo=pytorch)](https://pytorch.org)

![Tree Sitter](https://img.shields.io/badge/tree_sitter-0.23.0-7E8F31)

## Overview

This project focuses on translating x86-64 assembly back into C code using a

machine learning model trained on a dataset of C code snippets. Each snippet is

compiled with multiple optimization levels across different compilers, and the

resulting assembly code is tokenized for use in training.

## Dataset

- the model was trained on an augmented version of

  [this dataset](https://www.kaggle.com/datasets/shirshaka/c-code-snippets-and-their-labels)

- each snippet of C code is compiled (by default) with the first four

  optimization levels of GCC and Clang, yielding 8 unique assembly code snippets

  for each element in the initial dataset (totaling 2.5 million snippets)

If `kaggle` is in your path, the original dataset can be downloaded with:

```sh

kaggle datasets download -d shirshaka/c-code-snippets-and-their-labels && \

unzip -d dataset c-code-snippets-and-their-labels.zip

```

### Generation

- compilation is performed as needed when calling `DatasetIterator.take(n)`

  - compilation settings, including optimization levels and compiler choices are

    specified in the arguments to this method call

- the exact flags passed into the compilation subprocesses are specified in the

  `.compile()` methods in `compilation.py`

## Tokenization

C and assembly code snippets are tokenized semantically using the tree-sitter

library. Each token includes raw text paired with its symbolic identity, e.g.,

`(variable, 42)`.