Trans-Encoder: Unsupervised sentence-pair modelling through self- and mutual-distillations

Benchmark Model Rank Results
semantic-textual-similarity-on-sickTrans-Encoder-BERT-base-bi (unsup.)#12Spearman Correlation: 0.7276
semantic-textual-similarity-on-sickTrans-Encoder-BERT-large-cross (unsup.)#13Spearman Correlation: 0.7192
semantic-textual-similarity-on-sickTrans-Encoder-RoBERTa-large-cross (unsup.)#14Spearman Correlation: 0.7163
semantic-textual-similarity-on-sickTrans-Encoder-BERT-large-bi (unsup.)#15Spearman Correlation: 0.7133
semantic-textual-similarity-on-sickTrans-Encoder-BERT-base-cross (unsup.)#18Spearman Correlation: 0.6952
semantic-textual-similarity-on-sts-benchmarkTrans-Encoder-RoBERTa-large-cross (unsup.)#35Spearman Correlation: 0.867
semantic-textual-similarity-on-sts-benchmarkTrans-Encoder-RoBERTa-large-bi (unsup.)#36Spearman Correlation: 0.8655
semantic-textual-similarity-on-sts-benchmarkTrans-Encoder-BERT-large-bi (unsup.)#39Spearman Correlation: 0.8616
semantic-textual-similarity-on-sts-benchmarkTrans-Encoder-RoBERTa-base-cross (unsup.)#42Spearman Correlation: 0.8465
semantic-textual-similarity-on-sts-benchmarkTrans-Encoder-BERT-base-bi (unsup.)#45Spearman Correlation: 0.839
semantic-textual-similarity-on-sts12Trans-Encoder-RoBERTa-large-cross (unsup.)#7Spearman Correlation: 0.7828
semantic-textual-similarity-on-sts12Trans-Encoder-BERT-large-bi (unsup.)#8Spearman Correlation: 0.7819
semantic-textual-similarity-on-sts12Trans-Encoder-RoBERTa-base-cross (unsup.)#10Spearman Correlation: 0.7637
semantic-textual-similarity-on-sts12Trans-Encoder-BERT-base-bi (unsup.)#11Spearman Correlation: 0.7509
semantic-textual-similarity-on-sts13Trans-Encoder-BERT-large-bi (unsup.)#7Spearman Correlation: 0.8851
semantic-textual-similarity-on-sts13Trans-Encoder-BERT-large-cross (unsup.)#8Spearman Correlation: 0.8831
semantic-textual-similarity-on-sts13Trans-Encoder-RoBERTa-large-cross (unsup.)#9Spearman Correlation: 0.8831
semantic-textual-similarity-on-sts13Trans-Encoder-BERT-base-cross (unsup.)#11Spearman Correlation: 0.8559
semantic-textual-similarity-on-sts13Trans-Encoder-BERT-base-bi (unsup.)#12Spearman Correlation: 0.851
semantic-textual-similarity-on-sts14Trans-Encoder-RoBERTa-large-cross (unsup.)#9Spearman Correlation: 0.8194
semantic-textual-similarity-on-sts14Trans-Encoder-RoBERTa-large-bi (unsup.)#10Spearman Correlation: 0.8176
semantic-textual-similarity-on-sts14Trans-Encoder-BERT-large-bi (unsup.)#11Spearman Correlation: 0.8137
semantic-textual-similarity-on-sts14Trans-Encoder-RoBERTa-base-cross (unsup.)#12Spearman Correlation: 0.7903
semantic-textual-similarity-on-sts14Trans-Encoder-BERT-base-bi (unsup.)#13Spearman Correlation: 0.779
semantic-textual-similarity-on-sts15Trans-Encoder-RoBERTa-large-cross (unsup.)#6Spearman Correlation: 0.8863
semantic-textual-similarity-on-sts15Trans-Encoder-BERT-large-bi (unsup.)#7Spearman Correlation: 0.8816
semantic-textual-similarity-on-sts15Trans-Encoder-RoBERTa-base-cross (unsup.)#10Spearman Correlation: 0.8577
semantic-textual-similarity-on-sts15Trans-Encoder-BERT-base-bi (unsup.)#11Spearman Correlation: 0.8508
semantic-textual-similarity-on-sts15Trans-Encoder-BERT-base-cross (unsup.)#12Spearman Correlation: 0.8444
semantic-textual-similarity-on-sts16Trans-Encoder-RoBERTa-large-cross (unsup.)#7Spearman Correlation: 0.8503
semantic-textual-similarity-on-sts16Trans-Encoder-BERT-large-bi (unsup.)#9Spearman Correlation: 0.8481
semantic-textual-similarity-on-sts16Trans-Encoder-RoBERTa-base-cross (unsup.)#11Spearman Correlation: 0.8377
semantic-textual-similarity-on-sts16Trans-Encoder-BERT-base-bi (unsup.)#12Spearman Correlation: 0.8305