GraphCodeBERT: Pre-training Code Representations with Data Flow

Benchmark Model Rank Results
code-search-on-codesearchnetGraphCodeBERT#3Overall: 77.4Go: 84.1Ruby: 73.2Python: 87.9Java: 75.7…
type-prediction-on-manytypes4typescriptGraphCodeBERT#3Average Accuracy: 62.51Average Precision: 60.06Average Recall: 61.08…