Can LLMs Reason About Runtime Behavior? A Repository-Level Dynamic Benchmark
Large language models (LLMs) are increasingly used in coding tasks, but their ability to reason about code execution remains unclear. Existing repository-level QA benchmarks mainly evaluate static code understanding and often rely on LLM-based evaluation, while execution-reasoning benchmarks are mostly limited to snipp...
Hamed Taherkhani, Mohammad Abdollahi, Melika Sepidband et al.
· 0 citations