# The CodeInverter Suite: Structure- and Data-Aware Binary Decompilation with Efficient LLMs

> CodeInverter studies function-level LLM decompilation using control-flow graphs and memory-to-data mappings extracted with IDA/objdump. It presents a prompt workflow, training corpus and model family, evaluated through compilation, benchmark input/output tests and edit similarity.

- Authors: Peipei Liu; Jian Sun; Rongkang Sun; Li Chen; Zhaoteng Yan; Xiaoling Zhang; Dawei Wang; Dapeng Sun; Peizheng Zhang; Dan Li
- Venue: ASE 2026
- Status: Conference paper
- Publication / posting date: October 7, 2026
- Date source: Online and print publication date in ACM’s Crossref deposit; arXiv first submission is recorded separately.
- First arXiv submission: March 10, 2025
- arXiv: 2503.07215
- Reviewed arXiv version: v3
- DOI: 10.1145/3832783.3837554
- Sources reviewed: 2026-10-19
- Added to the catalog: 2026-10-20
- Original paper: <https://dl.acm.org/doi/10.1145/3832783.3837554>
- PDF: <https://arxiv.org/pdf/2503.07215v3>
- Code: <https://github.com/LiuPeiP-CS/CodeInverter>
- Project: <https://huggingface.co/CodeInverter>
- BibTeX: <https://awesomerea.org/papers/citations/paper-codeinverter-suite.bib>
- Catalog page: <https://awesomerea.org/papers/paper-codeinverter-suite/>
- Record JSON: <https://awesomerea.org/papers/paper-codeinverter-suite/record.json>
- Chinese version: [简体中文](https://awesomerea.org/zh/papers/paper-codeinverter-suite/index.md)

## Scope and limitations

Evaluation covers individual C functions under 32/64-bit GCC settings. Whole-program recovery, aggressive optimization and protected binaries remain limitations. The text reports one epoch (15 steps), while the pinned launch example sets one epoch and 100 warmup steps; the exact training schedule remains unresolved.

## Why it fits this collection

Adds structure- and data-informed LLM decompilation with separate compilation, execution-test and similarity evidence.

## Publication and review notes

The v3 title and ten-author order match ACM’s Crossref deposit. The repository README at 548e59c still cites v2/2025. Direct ACM access returned HTTP 403. Artifact snapshots are separately versioned. The training directory contains MIT and DeepSeek notices; the CIM-1.3B card declares CC BY 4.0.

## Primary sources

1. <https://api.crossref.org/works/10.1145/3832783.3837554>
2. <https://arxiv.org/abs/2503.07215v3>
3. <https://arxiv.org/pdf/2503.07215v3>
4. <https://arxiv.org/html/2503.07215v3>
5. <https://arxiv.org/abs/2503.07215v1>
6. <https://arxiv.org/abs/2503.07215v2>
7. <https://github.com/LiuPeiP-CS/CodeInverter/blob/548e59c5db07bd108659bf2bfb069f1cfd0c9298/README.md>
8. <https://zenodo.org/api/records/19351453>
9. <https://huggingface.co/CodeInverter/CIM-1.3b/raw/1f7745e19c5420a8c5896f5c31d6938f83454be4/README.md>
10. <https://github.com/LiuPeiP-CS/CodeInverter/blob/548e59c5db07bd108659bf2bfb069f1cfd0c9298/code4train/train/run_ms2decllm_train.sh>
11. <https://github.com/LiuPeiP-CS/CodeInverter/blob/548e59c5db07bd108659bf2bfb069f1cfd0c9298/code4train/LICENSE>
12. <https://github.com/LiuPeiP-CS/CodeInverter/blob/548e59c5db07bd108659bf2bfb069f1cfd0c9298/code4train/LICENSE-MODEL>

## About this entry

Reviewed against the paper and its primary sources. Cite the original authors using the linked paper or BibTeX.
Original editorial content: MIT. Linked projects, papers and assets retain their own rights and licenses.