arXiv Machine Learning

Training Language Model Agents to Find Vulnerabilities with CTF-Dojo

arXiv AI
Jun 9

MEnvAgent: Scalable Polyglot Environment Construction for Verifiable Software Engineering

arXiv:2601. 22859v3 Announce Type: replace-cross Abstract: The evolution of Large Language Model (LLM) agents for software engineering (SWE) is constrained by the scarcity of verifiable datasets, a bottleneck stemming from the complexity of constructing executable environments across diverse languages.

By Chuanzhe Guo, Jingjing Wu, Sijun He, Yang Chen, Zhaoqi Kuang, Shilong Fan, Bingjin Chen, Siqi Bao, Jing Liu, Hua Wu, Qingfu Zhu, Wanxiang Che, Haifeng Wang
arXiv AI
Oct 1

EngramBench: A Capability-Grounded Benchmark for Skill-Evolution Harnesses

EngramBench is a new benchmark designed to evaluate skill evolution in autonomous agents by focusing on genuine capability abstraction rather than solution copying. It includes 30 learning tasks and 13 unseen transfer tasks that require agents to manage complex, multi-hour development cycles with LLM‑simulated users. The study shows that while static skill banks cannot eliminate the need for precise code implementation, they effectively reduce redundant context and cut overall coding time by more than 55%.

By Zhixuan Tan, Pengjie Gu, Zhao Li, Yihan Hu, Xu He, Dong Li, Jianye Hao
arXiv Machine Learning
1d ago

Cyber-Zero: Training Cybersecurity Agents without Runtime

Cyber-Zero introduces the first runtime‑free framework for training cybersecurity agents by synthesizing high‑quality agent trajectories using publicly available CTF writeups and persona‑driven LLM simulation. The approach generates realistic, long‑horizon interaction sequences without requiring actual execution environments, enabling LLM‑based agents to achieve up to 11.4% absolute performance gains on three major CTF benchmarks. The best model, Cyber‑Zero‑32B, matches proprietary systems such as DeepSeek‑V3‑0324 and Claude‑3.5‑Sonnet while being more cost‑effective, demonstrating that runtime‑free trajectory synthesis can democratize state‑of‑the‑art cybersecurity agent development.

By Terry Yue Zhuo, Dingmin Wang, Hantian Ding, Varun Kumar, Zijian Wang
arXiv AI
Oct 12

ProgramGym: Scaling Training Environments for Long-Horizon Program Reproduction Agents

ProgramGym introduces a scalable framework for training long-horizon program reproduction agents, comprising Program-Env and Program-Agent components. Program-Env builds verifiable training environments from repositories via a feedback loop that repairs sandbox builds and improves black-box tests, yielding 1,648 environments across C, C++, Go, and Rust. Program-Agent employs a role-isolated workflow of analysis, implementation, and review with revision management, boosting performance on ProgramBench and raising Qwen3.8-27B’s accuracy from 34.87% to 42.84% and surpassing flagship models with fine-tuning.

By Chengyi Yang, Renxu Bao, Jue Hou, Zongpei Teng, Ziheng Wang, Yujie Lin, Chengsong Huang, Junnan Zhu, Delai Qiu, Yining Wang, Shengping Liu, Fei Long, Jinsong Su
arXiv AI
Oct 26

daVinci-Env: Open SWE Environment Synthesis at Scale

arXiv:2603.13023v3 Announce Type: replace-cross Abstract: Training capable software engineering (SWE) agents demands large-scale, executable, and verifiable environments that provide dynamic feedback...

By Dayuan Fu, Shenyu Wu, Yunze Wu, Zerui Peng, Yaxing Huang, Jie Sun, Ji Zeng, Mohan Jiang, Lin Zhang, Yukun Li, Jiarui Hu, Liming Liu, Jinlong Hou, Pengfei Liu
arXiv AI
3d ago

Together Compiler: Continual Compilation for AI Inference

Together Compiler (TKCC) is a continual agent‑driven optimization system for AI inference that addresses the trade‑off between measurement throughput and fidelity. It introduces a Tiered Persistent Engine to separate engine launch overhead from forward‑pass computation and a cost model to seed parallelism configurations. TKCC achieves up to 1.7× speed‑ups over hand‑written Megakernels on Llama 1B and up to 2.6× higher TPS on frontier models across hardware platforms, while also delivering state‑of‑the‑art batch‑size‑1 performance.

By Reyna Abhyankar, Dylan Lim, Drew Wadsworth, Venkata Hasith Vattikuti, YuTong Julia Zhang, Jung Hwan Heo, Taeksang Kim, Kesavan Ramakrishnan, Evangelene Stanley, Xinyu Fang, Daniel Y. Fu, Simran Arora