List your company

Code Generation on HumanEval (Correction Uplift %)

62.5Correction Uplift (%)ROSA
12.465625.455338.44551.4347Sep 27, 2025

Evaluation results

MethodLinks
ROSAModel=Qwen3-8B2025.0962.5
BaselineModel=Qwen3-8B2025.0947.83
ROSAModel=Qwen2.5-7B-Instruct2025.0940
ROSAModel=DeepSeek-R1-Distill-Llama-8B2025.0931.97
ROSAModel=Qwen3-0.6B2025.0931.01
BaselineModel=Qwen2.5-7B-Instruct2025.0928.57
ROSAModel=Qwen2.5-0.5B-Instruct2025.0926.09
BaselineModel=Qwen3-0.6B2025.0919.33
BaselineModel=DeepSeek-R1-Distill-Llama-8B2025.0915.49
BaselineModel=Qwen2.5-0.5B-Instruct2025.0914.39