MNLI
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
MNLI Hypothesis-only Bias (in-distribution)
84.2Accuracy
13
MNLI Syntactic Bias (in-distribution)
84.3Accuracy
13
MNLI (all combined)
85.98Accuracy
12
MNLI-m (dev)
90.6Accuracy
12
MNLI bias-free (test)
76Worst Group Accuracy
10
MNLI (test)
100AuROC
10
MNLI
89.5Accuracy
10
MNLI source domains (out-of-domain)
0.683ROC AUC
10
MNLI source domains (in-domain)
0.699ROC AUC
10
MNLI
86.54Accuracy
10
MNLI
0.95Kendall Correlation
10
MNLI matched mismatched (dev)
84Accuracy (Matched)
10
MNLI Matched (test)
90.2Accuracy
9
MNLI mismatched (val)
75.01Accuracy
9
MNLI Mismatched Hard (test)
79.28Accuracy
9
MNLI Matched Hard (test)
80.03Accuracy
9
MNLI GLUE matched mismatched average (test)
84.8Accuracy
9
MNLI
79.15MNLI (m) Accuracy
8
MNLI TR
36.42Score
8
MNLI mm
44.1ASR
8
MNLI m
38.2ASR
8
MNLI
33.1ASR
8
MNLI Matched Hard (dev)
83.54Accuracy
8
MNLI Out-of-Domain (test)
79.6Accuracy
8
adv-MNLI TextFooler-RoBERTa
65.6Accuracy (Matched)
8