# SLaDe: A Portable Small Language Model Decompiler for Optimized Assembly

> SLaDe translates function-level assembly into C with a 200M-parameter encoder–decoder model, code-specific tokenization and PsycheC type completion. It studies x86 and ARM at -O0 and -O3, combining generated candidates with input/output checks. The input is assembly text, not raw machine bytes; disassembly is outside the method.

- Authors: Jordi Armengol-Estapé; Jackson Woodruff; Chris Cummins; Michael F. P. O'Boyle
- Venue: IEEE/ACM CGO 2024
- Status: Conference paper
- Publication / posting date: March 2, 2024
- Date source: IEEE’s Crossref deposit gives published/print date 2024-03-02. Edinburgh’s publication record instead gives 2024-02-28, also the DOI-creation date. This entry uses the publisher-deposited print date and does not infer it solely from the conference schedule.
- First arXiv submission: May 21, 2023
- arXiv: 2305.12520
- Reviewed arXiv version: v3
- DOI: 10.1109/CGO57630.2024.10444788
- Sources reviewed: 2026-10-18
- Added to the catalog: 2026-10-18
- Original paper: <https://ieeexplore.ieee.org/document/10444788>
- PDF: <https://arxiv.org/pdf/2305.12520v3>
- Code: <https://zenodo.org/records/10205121>
- BibTeX: <https://awesomerea.org/papers/citations/paper-slade.bib>
- Catalog page: <https://awesomerea.org/papers/paper-slade/>
- Record JSON: <https://awesomerea.org/papers/paper-slade/record.json>
- Chinese version: [简体中文](https://awesomerea.org/zh/papers/paper-slade/index.md)

## Scope and limitations

Evaluation covers about 4,000 ExeBench functions and a 112-program synthesis suite. Beam size 5 selects the first candidate passing finite I/O tests; generated functions are inserted into the original calling context. Passing these tests is not whole-program or all-input equivalence. Edit similarity is a readability proxy. Longer functions, external calls and incorrect type/operator predictions remain failure modes; the model has 1,024-position source and target windows. Historical baseline comparisons have different coverage and do not establish current rankings.

## Why it fits this collection

Adds type-inference-assisted neural decompilation with explicit execution-based evaluation, complementing symbol recovery and later LLM decompilation studies.

## Publication and review notes

Reviewed arXiv v3 dated 2024-02-15, the accepted CGO manuscript; publication metadata establishes the final venue and pages. Paper reference 65 links the Zenodo artifact. Its separate LICENSE restricts included code/data/models to paper reproducibility and calls it nonproduction research code. Appendix prerequisites include Python 3.8, PyTorch 1.10.2, Transformers 4.22.2, PsycheC, Haskell Stack and an AArch64 GCC toolchain. BTC weights are not redistributed and ChatGPT outputs may change.

## Primary sources

1. <https://ieeexplore.ieee.org/document/10444788>
2. <https://api.crossref.org/works/10.1109/CGO57630.2024.10444788>
3. <https://www.research.ed.ac.uk/en/publications/slade-a-portable-small-language-model-decompiler-for-optimized-as/>
4. <https://conf.researchr.org/details/cgo-2024/cgo-2024-main-conference/17/SLaDe-A-Portable-Small-Language-Model-Decompiler-for-Optimized-Assembly>
5. <https://arxiv.org/abs/2305.12520>
6. <https://arxiv.org/pdf/2305.12520v3>
7. <https://zenodo.org/records/10205121>
8. <https://zenodo.org/api/records/10205121>
9. <https://zenodo.org/api/records/10205121/files/LICENSE/content>

## 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.