MVP: Multi-task Supervised Pre-training for Natural Language Generation

Benchmark Model Rank Results
natural-language-inference-on-qnli-glue-devMVP#2Accuracy: 95.10
paraphrase-identification-on-mrpc-glue-devMVP#2F1: 92.07
natural-language-inference-on-multinli-devMVP#7Matched: 89.70
semantic-textual-similarity-on-sts-b-glue-devMVP#7Pearson/Spearman Correlation: 91.1
linguistic-acceptability-on-cola-glue-devMVP#9Matthews Correlation: 59.87
natural-language-understanding-on-glue-devMVP#9GLUE (dev): 85.88
natural-language-inference-on-rte-glue-devMVP#10Accuracy: 82.87
text-summarization-on-x-sumMVP+S#19ROUGE-L: 37.50
abstractive-text-summarization-on-cnn-daily-mailMVP+S#43ROUGE-L: 41.21
sentiment-analysis-on-sst-2-binary-classificationMVP#86Dev Accuracy: 96.43