List your company

Code Debug on InfiniteBench Code Debug

74.37AccuracyFull Prompt
58.520462.635266.7570.8648Mar 3, 2026

Evaluation results

MethodLinks
Full PromptTarget Model=DeepSeek-R1-0528, Keep rate=–2026.0374.37
SP w/ Llama-3.1-8B-InsTarget Model=DeepSeek-R1-0528, Draft Model=Llama-3.1-8B-Ins, Keep rate=30%2026.0370.3
SP w/ Llama-3.1-8B-InsTarget Model=DeepSeek-R1-0528, Draft Model=Llama-3.1-8B-Ins, Keep rate=25%2026.0368.02
Full PromptTarget Model=DeepSeek-V3.1, Keep rate=–2026.0367.51
SP w/ Llama-3.1-8B-InsTarget Model=DeepSeek-V3.1, Draft Model=Llama-3.1-8B-Ins, Keep rate=20%2026.0364.72
SP w/ Llama-3.1-8B-InsTarget Model=DeepSeek-R1-0528, Draft Model=Llama-3.1-8B-Ins, Keep rate=15%2026.0362.44
SP w/ Llama-3.1-8B-InsTarget Model=DeepSeek-V3.1, Draft Model=Llama-3.1-8B-Ins, Keep rate=15%2026.0359.13