AMBER
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
AMBER
89.4Accuracy
10
AMBER Discrimination 1.0 (test)
76.7Accuracy
10
AMBER
12.5PPL1
9
AMBER Discriminative
90.3F1 Score
9
AMBER out-of-distribution (OOD)
0.861AUC
8
AMBER AttackVLM (adversarially perturbed)
24.4CHAIR
6
AMBER Generative Subset benign setting
3.2CHAIR Score
6
AMBER
1Total Time
5
AMBER
84.3Accuracy
4
Amber 1.2T tokens
4.28BPD
4
AMBER (test)
7.28Accuracy
2