DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence

Benchmark Model Rank Results
code-generation-on-appsdeepseek-ai/deepseek-coder-6.7b-instruct#5Introductory Pass@1: 33.80Interview Pass@1: 19.70
code-generation-on-mbppGPT-4 (few-shot)#16Accuracy: 80
code-generation-on-mbppGPT-3.5 Turbo (few-shot)#19Accuracy: 70.8
code-generation-on-mbppDeepSeek-Coder-Instruct 33B (few-shot)#20Accuracy: 70
code-generation-on-mbppDeepSeek-Coder-Base 33B (few-shot)#28Accuracy: 66
code-generation-on-mbppDeepSeek-Coder-Instruct 6.7B (few-shot)#30Accuracy: 65.4
code-generation-on-mbppDeepSeek-Coder-Base 6.7B (few-shot)#38Accuracy: 60.6
code-generation-on-mbppDeepSeek-Coder-Instruct 1.3B (few-shot)#51Accuracy: 49.4
code-generation-on-mbppDeepSeek-Coder-Base 1.3B (few-shot)#63Accuracy: 46.2