Text and Code Embeddings by Contrastive Pre-Training

Benchmark Model Rank Results
code-search-on-codesearchnetcpt-code M#1Overall: 93.5Go: 97.5Ruby: 85.5Python: 99.9Java: 94.4
code-search-on-codesearchnetcpt-code S#2Overall: 93.4Go: 97.7Ruby: 86.3Python: 99.8Java: 94.0
passage-ranking-on-ms-marcoFine-tuned SOTA#1MRR@10: 44.3
passage-ranking-on-ms-marcocpt-text XL#2MRR@10: 22.7
passage-ranking-on-ms-marcocpt-text L#3MRR@10: 21.5
passage-ranking-on-ms-marcoBM25#4MRR@10: 18.4