Self-Execution Simulation Improves Coding Models

Open paper

Benchmark	Model	Rank	Results
code-generation-on-mbpp	Qwen2.5-7B + NLEX	–	Accuracy: 63.2