Natural Language Understanding on GLUE v2 (dev)
93.3SST-2 ScoreTISA
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| TISABackbone=ALBERT base v2, Position Embeddings=Conventional, S (TISA Kernels)=1, Evaluation Protocol=Median over 5 runs2021.06 | 93.3 | 84.1 | 88 | 90.4 | 57 | 90.1 | 91.7 | 71.1 | — | — | |
| TISABackbone=ALBERT base v2, Position Embeddings=Conventional, S (TISA Kernels)=3, Evaluation Protocol=Median over 5 runs2021.06 | 93.1 | 84.4 | 88.3 | 90 | 56.5 | 89 | 91.4 | 73.6 | — | — | |
| TISABackbone=ALBERT base v2, Position Embeddings=Conventional, S (TISA Kernels)=5, Evaluation Protocol=Median over 5 runs2021.06 | 93.1 | 84.8 | 88.3 | 90.4 | 58.5 | 90.1 | 91.6 | 73.6 | — | — | |
| ALBERT base v2 BaselineBackbone=ALBERT base v2, Position Embeddings=Conventional, S (TISA Kernels)=0, Evaluation Protocol=Median over 5 runs2021.06 | 92.9 | 83.8 | 88.2 | 90.3 | 57.2 | 89.6 | 91.6 | 72.9 | — | — |