ResearchBenchmarksNatural Language Understanding on GLUE (val) - Subset (SST-2, MRPC, MNLI, STS-B)Follow68.69SST-2 AccuracyMLM-only67.504467.812268.1268.4278Apr 16, 2026Evaluation ResultsMethodMethodLinksSST-2 AccuracyMRPC F1 ScoreMNLI AccuracySTS-B Spearman CorrelationMLM-onlyEvaluation protocol=Li...Evaluation protocol=Linear-probe, Encoder state=Frozen, Pooling strategy=Mean-pooled2026.0468.6959.8451.360.283HybridEvaluation protocol=Li...Evaluation protocol=Linear-probe, Encoder state=Frozen, Pooling strategy=Mean-pooled2026.0467.5563.0950.820.281