On the Sentence Embeddings from Pre-trained Language Models

Benchmark Model Rank Results
semantic-textual-similarity-on-sickBERTbase-flow (NLI)#21Spearman Correlation: 0.6544
semantic-textual-similarity-on-sts-benchmarkBERTlarge-flow (target)#53Spearman Correlation: 0.7226
semantic-textual-similarity-on-sts12BERTlarge-flow (target)#18Spearman Correlation: 0.6520
semantic-textual-similarity-on-sts13BERTlarge-flow (target)#21Spearman Correlation: 0.7339
semantic-textual-similarity-on-sts14BERTlarge-flow (target)#20Spearman Correlation: 0.6942
semantic-textual-similarity-on-sts15BERTlarge-flow (target)#20Spearman Correlation: 0.7492
semantic-textual-similarity-on-sts16BERTlarge-flow (target)#16Spearman Correlation: 0.7763