MT-DNN-SMART

Reported on 6 benchmarks across 4 tasks · 1 paper · 3 SOTA

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

Natural Language Processing6 results

Semantic Textual SimilarityonSTS Benchmark
Pearson Correlation· 2019-11-08
0.929
SOTA
SMART: Robust and Efficient Fine-Tuning for Pre-trained Natural Language Models through Principled Regularized Optimization arXiv:1911.03437
Semantic Textual SimilarityonSTS Benchmark
Spearman Correlation· 2019-11-08
0.925
best: 0.931 (Mnet-Sim)
SOTA
SMART: Robust and Efficient Fine-Tuning for Pre-trained Natural Language Models through Principled Regularized Optimization arXiv:1911.03437
Natural Language UnderstandingonGLUE
Average· 2019-11-08
89.9
SOTA
SMART: Robust and Efficient Fine-Tuning for Pre-trained Natural Language Models through Principled Regularized Optimization arXiv:1911.03437
Natural Language InferenceonMultiNLI
Accuracy· 2019-11-08
85.7
SMART: Robust and Efficient Fine-Tuning for Pre-trained Natural Language Models through Principled Regularized Optimization arXiv:1911.03437
Semantic Textual SimilarityonMRPC
F1· 2019-11-08
91.7
best: 92.5 (T5-3B)
SMART: Robust and Efficient Fine-Tuning for Pre-trained Natural Language Models through Principled Regularized Optimization arXiv:1911.03437
Sentiment AnalysisonSST-2 Binary classification
Accuracy· 2019-11-08
97.5
SMART: Robust and Efficient Fine-Tuning for Pre-trained Natural Language Models through Principled Regularized Optimization arXiv:1911.03437