CodeT: Code Generation with Generated Tests

Benchmark Model Rank Results
code-generation-on-appscode-davinci-002 175B (CodeT)#4Introductory Pass@1: 47.3%Interview Pass@1: 14.3%
code-generation-on-appscode-davinci-002 175B#6Introductory Pass@1: 31.92
code-generation-on-mbppcode-davinci-002 175B + CodeT#23Accuracy: 67.7
code-generation-on-mbppcode-davinci-001 175B + CodeT#34Accuracy: 61.9
code-generation-on-mbppcode-cushman-001 12B (CodeT)#42Accuracy: 55.4
code-generation-on-mbppCodeGen-Mono 16B + CodeT#49Accuracy: 49.5
code-generation-on-mbppInCoder 6.7B + CodeT#77Accuracy: 34.4