DeBERTa (large)

Reported on 5 benchmarks across 4 tasks · 1 paper · 2 SOTA

Note: results are matched by exact model name. Different papers may use the same name for different model variants.

Natural Language Processing5 results

Semantic Textual SimilarityonSTS Benchmark
Accuracy· 2020-06-05
92.5
SOTA
DeBERTa: Decoding-enhanced BERT with Disentangled Attention arXiv:2006.03654
Linguistic AcceptabilityonCoLA Dev
Accuracy· 2020-06-05
69.5
best: 88.6 (En-BERT + TDA)
SOTA
DeBERTa: Decoding-enhanced BERT with Disentangled Attention arXiv:2006.03654
Natural Language InferenceonMultiNLI
Matched· 2020-06-05
91.1
best: 92.6 (Turing NLR v5 XXL 5.4B (fine-tuned))
DeBERTa: Decoding-enhanced BERT with Disentangled Attention arXiv:2006.03654
Natural Language InferenceonMultiNLI
Mismatched· 2020-06-05
91.1
best: 92.4 (Turing NLR v5 XXL 5.4B (fine-tuned))
DeBERTa: Decoding-enhanced BERT with Disentangled Attention arXiv:2006.03654
Sentiment AnalysisonSST-2 Binary classification
Accuracy· 2020-06-05
96.5
best: 97.5 (T5-11B)
DeBERTa: Decoding-enhanced BERT with Disentangled Attention arXiv:2006.03654