LLMs for Code

We study large language models as tools for coding and software development. We focus on the issues of constrained decoding and secure code generation, approaching the latter from the perspectives of evaluation, adversarial attacks that elicit insecure code generations, and developing methods to improve this capability.

Startups

LogicStar

LogicStar develops autonomous AI for software engineering.

Publications

2025

Constrained Decoding of Diffusion LLMs with Context-Free Grammars

Niels Mündler, Jasper Dekoninck, Martin Vechev

arXiv 2025

Website
Paper
Code

BaxBench: Can LLMs Generate Secure and Correct Backends?

Mark Vero, Niels Mündler, Victor Chibotaru, Veselin Raychev, Maximilian Baader, Nikola Jovanović, Jingxuan He, Martin Vechev

ICML 2025 Spotlight

Website
Paper
Code

Black-Box Adversarial Attacks on LLM-Based Code Completion

Slobodan Jenko*, Niels Mündler*, Jingxuan He, Mark Vero, Martin Vechev

ICML 2025 * Equal contribution

Paper

Type-Constrained Code Generation with Language Models

Niels Mündler^†, Jingxuan He^†, Hao Wang, Koushik Sen, Dawn Song, Martin Vechev

PLDI 2025 † Co-leadership

Slides
Talk
Paper
Code

Automated Benchmark Generation for Repository-Level Coding Tasks

Konstantinos Vergopoulos*, Mark Niklas Müller*, Martin Vechev

ICML 2025 * Equal contribution

Paper
Code

2024

SWT-Bench: Testing and Validating Real-World Bug-Fixes with Code Agents

Niels Mündler, Mark Niklas Müller, Jingxuan He, Martin Vechev

NeurIPS 2024

Website
Paper
Code

Instruction Tuning for Secure Code Generation

Jingxuan He*, Mark Vero*, Gabriela Krasnopolska, Martin Vechev

ICML 2024 * Equal contribution

Paper
Code

2023

Large Language Models for Code: Security Hardening and Adversarial Testing

Jingxuan He, Martin Vechev

ACM CCS 2023 Distinguished Paper Award

Slides
Paper
Code