CoVer at EMNLP 2026: Training Code LLMs to Generate and Self-Verify Simultaneously
algo_diver · x · 2026-10-26
- Poster at EMNLP 2026 (Budapest) presents CoVer, a Findings paper on training one model to both generate code and reliably self-assess it, with generate/verify roles plus GRPO-style updates.
- Reports +32.35 average gain over base models, +8.93 over a self-breaking baseline, Pass@8 of 89.44 across HumanEval+, LiveCodeBench, BigCodeBench and CodeForces.
- Verification replays buffered code and runs available tests; self-rewarding works without labels or a separate judge model.
- Caveats: labels reliable only for code generation; results shown on Qwen models.
More from Research
- RCT: ChatGPT temporarily boosts consultants' data science skills, gains vanish without it — EricTopol · 2026-10-26
- FocusVLA routes robot camera views by predicted interaction intent across task stages — iscas-cloud · 2026-10-26
- Transformers' biggest audio overhaul: unified AudioProcessor architecture — LysandreJik · 2026-10-26
- Free 195-page primer: The Little Book of Generative AI Foundations — caglar_ee · 2026-10-26
- SPADER from Zhejiang and Peking University Tops All Four Multi-Answer QA Benchmarks, Accepted at EMNLP — jiqizhixin · 2026-10-26
- Purdue's GPS-free autonomous drone tracks runners through dense forests using on-chip LiDAR AI — KordingLab · 2026-10-26