Skip to Main Content
Table 7: 

GLUE development set results for different architectures for incorporating semantic information. The settings and metrics are identical to Table 3a. All models use the base size variant.

MNLI
ModelsCoLAMRPCRTESST-2STS-BQNLIQQPID.OOD.Avg.
RoBERTa 63.1 90.1 79.0 94.6 91.0 93.0 91.8 87.7 87.3 86.4 
 
GCN 65.2 90.2 80.2 94.8 91.1 92.9 91.8 87.8 87.7 86.8 
GAT 63.4 90.0 79.4 94.7 91.2 92.9 91.8 87.7 87.6 86.5 
Hidden 64.2 90.2 79.7 94.5 91.0 92.8 91.8 87.1 86.7 86.4 
Scaffold 62.5 90.5 71.1 94.3 91.0 92.6 91.7 87.7 87.6 85.5 
 
SIFT 64.8 90.5 81.0 95.1 91.3 93.2 91.9 87.9 87.7 87.0 
SIFT-Light 64.1 90.3 80.6 94.7 91.2 92.8 91.7 87.7 87.6 86.7 
MNLI
ModelsCoLAMRPCRTESST-2STS-BQNLIQQPID.OOD.Avg.
RoBERTa 63.1 90.1 79.0 94.6 91.0 93.0 91.8 87.7 87.3 86.4 
 
GCN 65.2 90.2 80.2 94.8 91.1 92.9 91.8 87.8 87.7 86.8 
GAT 63.4 90.0 79.4 94.7 91.2 92.9 91.8 87.7 87.6 86.5 
Hidden 64.2 90.2 79.7 94.5 91.0 92.8 91.8 87.1 86.7 86.4 
Scaffold 62.5 90.5 71.1 94.3 91.0 92.6 91.7 87.7 87.6 85.5 
 
SIFT 64.8 90.5 81.0 95.1 91.3 93.2 91.9 87.9 87.7 87.0 
SIFT-Light 64.1 90.3 80.6 94.7 91.2 92.8 91.7 87.7 87.6 86.7 
Close Modal

or Create an Account

Close Modal
Close Modal