GOAT-Bench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
GOAT-Bench unseen (val)
90.6SR
37
GOAT-Bench 1.0 (test)
75.1Harmfulness Accuracy
32
GOAT-Bench unseen (val)
71.4Success Rate (SR)
21
GOAT-Bench
73.7SR
21
GOAT-Bench 1/10-scale subset (val-unseen)
72.8Success Rate
13
GOAT-Bench Seen (val)
49SPL
13
GOAT-Bench In-Domain
88.5Racism F1
11
GOAT-Bench unseen (val)
0.465s-SR
10
GOAT-Bench unseen (val)
62.7SR
9
GOAT-Bench
69.1Success Rate (SR)
8
GOAT-Bench (Out-Of-Domain)
87.1Racism F1
7
GOAT-BENCH Strict Physical Regime full 2,669-subtask (val_unseen)
55.6Subtask SR
6
Full GOAT-Bench Unseen (val)
54.3SR
6
Full GOAT-Bench Synonyms (val)
58.4Success Rate (SR)
6
Full GOAT-Bench Seen (val)
56.7SR
6
GOAT-Bench Seen-Synonyms (val)
66.8SR
6
GOAT-Bench (val-seen)
34.5SR
4
GOAT-Bench seen (val)
42.2Success Rate (SR)
4
GOAT-Bench evaluation 278 subtasks
72.41SR
4
GOAT-Bench full 2,780 subtasks (val)
64.14Success Rate (SR)
4
GOAT-Bench unseen (val)
45SR (%)
4
GOAT-BENCH Relaxed Continuous Regime full 2,669-subtask (val_unseen)
62.9Subtask SR
1