Natural Language Understanding on GLUE (Full Suite)
63.1CoLA ScoreACL
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| ACLModel=RoBERTa, Method=ACL, #Flops=6.7G2026.02 | 63.1 | 86.8 | 86.9 | 90.8 | 92.3 | 90.2 | 75.3 | 94.2 | |
| RoBERTa (CE+SCL)Model=RoBERTa, Method=CE+SCL, #Flops=6.7G2026.02 | 62.7 | 87.1 | 86.7 | 90.7 | 92.1 | 89.5 | 74.2 | 93.6 | |
| RoBERTa (CE)Model=RoBERTa, Method=CE, #Flops=6.7G2026.02 | 62.4 | 86.9 | 86.5 | 90.6 | 91.8 | 90.3 | 73.6 | 93.7 | |
| ACLModel=BERT, Method=ACL, #Flops=6.6G2026.02 | 58.5 | 83 | 83.7 | 89.5 | 89.7 | 88.9 | 66.2 | 92.7 | |
| BERT (CE+SCL)Model=BERT, Method=CE+SCL, #Flops=6.6G2026.02 | 57.9 | 82.6 | 82.7 | 88.7 | 89.7 | 88.5 | 65.8 | 92.1 | |
| BERT (CE)Model=BERT, Method=CE, #Flops=6.6G2026.02 | 57.3 | 82.9 | 83.4 | 89.6 | 89.5 | 88.7 | 65.3 | 92.8 | |
| TinyBERT (6L)Model=TinyBERT (6L), Method=-, #Flops=3.3G2026.02 | 46.3 | 83.6 | 83.8 | 87.2 | 90.3 | 89.1 | 68.6 | 91.9 | |
| ACL-CLModel=ACL-RoBERTa-6L, Method=ACL-CL, #Flops=3.3G2026.02 | 46.2 | 83.3 | 83.5 | 87.5 | 90.5 | 88.9 | 69.5 | 92.3 | |
| DistilBERT (6L)Model=DistilBERT (6L), Method=-, #Flops=3.3G2026.02 | 46.1 | 82.1 | 82 | 86.2 | 88.3 | 88.5 | 64.1 | 91.3 | |
| HeadPruneModel=HeadPrune, Method=-, #Flops=4.7G2026.02 | 45.7 | 78 | 79.7 | 80.2 | 86.1 | 84.7 | 64.5 | 91 | |
| RoBERTa-6LModel=RoBERTa-6L, Method=-, #Flops=3.4G2026.02 | 45.6 | 82.5 | 83.9 | 86.9 | 89.8 | 88.6 | 67.8 | 91.6 | |
| BERT-of-TheseusModel=BERT-of-Theseus, Method=-, #Flops=3.3G2026.02 | 44.7 | 81.4 | 81.9 | 85.3 | 88.1 | 88.6 | 66.2 | 91.5 | |
| ACL-CLModel=ACL-BERT-6L, Method=ACL-CL, #Flops=3.3G2026.02 | 44.7 | 82 | 82.1 | 85.1 | 88.5 | 87.7 | 63.6 | 90.8 | |
| LayerDrop-6LModel=LayerDrop-6L, Method=-, #Flops=3.3G2026.02 | 43.7 | 82.3 | 82.4 | 86.5 | 89.3 | 89.1 | 67.3 | 91.8 | |
| BERT-6LModel=BERT-6L, Method=-, #Flops=3.3G2026.02 | 43.2 | 81.4 | 81.6 | 84.9 | 87.9 | 87.2 | 61 | 90.3 |