CodeBERT: A Pre-Trained Model for Programming and Natural Languages

Benchmark Model Rank Results
code-documentation-generation-onCodeBERT (MLM)#2Smoothed BLEU-4: 15.48
code-documentation-generation-onCodeBERT (MLM+RTD)#3Smoothed BLEU-4: 15.41
code-documentation-generation-onpre-train w/ code only#4Smoothed BLEU-4: 15.12
code-documentation-generation-onRoBERTa#5Smoothed BLEU-4: 14.92
code-documentation-generation-onTransformer#6Smoothed BLEU-4: 13.44
code-documentation-generation-onseq2seq#7Smoothed BLEU-4: 13.04
code-documentation-generation-on-1CodeBERT (MLM+RTD)#2Smoothed BLEU-4: 14.56
code-documentation-generation-on-1CodeBERT (MLM)#3Smoothed BLEU-4: 13.59
code-documentation-generation-on-1RoBERTa#4Smoothed BLEU-4: 13.2
code-documentation-generation-on-1pre-train w/ code only#5Smoothed BLEU-4: 13.07
code-documentation-generation-on-1CodeBERT (RTD)#6Smoothed BLEU-4: 12.72
code-documentation-generation-on-1Transformer#7Smoothed BLEU-4: 12.57
code-documentation-generation-on-1seq2seq#8Smoothed BLEU-4: 11.42
code-documentation-generation-on-2CodeBERT (MLM)#1Smoothed BLEU-4: 26.79
code-documentation-generation-on-2CodeBERT (MLM+RTD)#2Smoothed BLEU-4: 26.66
code-documentation-generation-on-2pre-train w/ code only#3Smoothed BLEU-4: 26.39
code-documentation-generation-on-2RoBERTa#4Smoothed BLEU-4: 26.09
code-documentation-generation-on-2CodeBERT (RTD)#5Smoothed BLEU-4: 26.02
code-documentation-generation-on-2seq2seq#6Smoothed BLEU-4: 23.48
code-documentation-generation-on-3CodeBERT (MLM+RTD)#2Smoothed BLEU-4: 21.32
code-documentation-generation-on-3CodeBERT (MLM)#3Smoothed BLEU-4: 21
code-documentation-generation-on-3pre-train w/ code only#4Smoothed BLEU-4: 20.71
code-documentation-generation-on-3CodeBERT (RTD)#5Smoothed BLEU-4: 20.25
code-documentation-generation-on-3RoBERTa#6Smoothed BLEU-4: 19.9
code-documentation-generation-on-3seq2seq#7Smoothed BLEU-4: 18.4
code-documentation-generation-on-3Transformer#8Smoothed BLEU-4: 18.25
code-documentation-generation-on-4CodeBERT (MLM+RTD)#2Smoothed BLEU-4: 8.46
code-documentation-generation-on-4CodeBERT (MLM)#3Smoothed BLEU-4: 7.95
code-documentation-generation-on-4Transformer#4Smoothed BLEU-4: 7.87
code-documentation-generation-on-4pre-train w/ code only#5Smoothed BLEU-4: 7.36
code-documentation-generation-on-4RoBERTa#6Smoothed BLEU-4: 7.26
code-documentation-generation-on-4seq2seq#7Smoothed BLEU-4: 6.96
code-documentation-generation-on-5Transformer#1Smoothed BLEU-4: 25.61
code-documentation-generation-on-5CodeBERT (MLM+RTD)#3Smoothed BLEU-4: 9.54
code-documentation-generation-on-5CodeBERT (RTD)#4Smoothed BLEU-4: 8.73
code-documentation-generation-on-5CodeBERT (MLM)#5Smoothed BLEU-4: 8.51
code-documentation-generation-on-5pre-train w/ code only#6Smoothed BLEU-4: 8.3
code-documentation-generation-on-5seq2seq#7Smoothed BLEU-4: 6.88
code-documentation-generation-on-5RoBERTa#8Smoothed BLEU-4: 5.72
code-search-on-codesearchnetCodeBERT#6Overall: 76.0Go: 69.3Ruby: 70.6Python: 84.0Java: 86.8
type-prediction-on-manytypes4typescriptCodeBERT#4Average Accuracy: 61.72Average Precision: 59.34Average Recall: 59.80