ResearchBenchmarksCode Generation on HumanEval (Accuracy, Steps, TPS, Speedup)Follow42.68AccuracyWINO+37.604838.922440.2441.5576May 16, 2026Evaluation ResultsMethodMethodLinksAccuracyStepsStep Reduction FactorTPSTPS Speedup FactorWINO+Backbone=LLaDA-8B-Inst...Backbone=LLaDA-8B-Instruct, Shot=0-shot2026.0542.6882.713.137.82.6WINOBackbone=LLaDA-8B-Inst...Backbone=LLaDA-8B-Instruct, Shot=0-shot2026.0542.0793.322.7437.192.56LLaDABackbone=LLaDA-8B-Inst...Backbone=LLaDA-8B-Instruct, Shot=0-shot2026.0537.8256114.521