ResearchBenchmarksCode Generation on HumanEval, HumanEval+, MBPP, MBPP+ (test)Follow55.5HumanEval ScoreHybrid KD52.58853.34454.154.856May 25, 2026Evaluation ResultsMethodMethodLinksHumanEval ScoreHumanEval+ ScoreMBPP ScoreMBPP+ ScoreAverage ScoreHybrid KDSetting=Qwen2.5-Coder-...Setting=Qwen2.5-Coder-7B→1.5B2026.0555.550.661.452.655Hard KDSetting=Qwen2.5-Coder-...Setting=Qwen2.5-Coder-7B→1.5B2026.0554.35060.352.154.2Soft KDSetting=Qwen2.5-Coder-...Setting=Qwen2.5-Coder-7B→1.5B2026.0552.747.959.652.453.1