Unsupervised Cross-lingual Representation Learning at Scale

Benchmark Model Rank Results
semantic-textual-similarity-on-sts-b-glue-devXLM-R#6Pearson/Spearman Correlation: 91.2
paraphrase-identification-on-mrpc-glue-devXLM-R#7F1: 89.5
natural-language-inference-on-multinli-devXLM-R#10Matched: 88.9
natural-language-inference-on-qnli-glue-devXLM-R#10Accuracy: 93.8
sentiment-analysis-on-sst-2-binary-classificationXLM-R#90Dev Accuracy: 95.0