Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Grokking Generalization on Modular Addition (train/test)
Loading...
14
Grokked Count
SupCon-wrong
-0.56
3.22
7
10.78
Jul 5, 2026
Grokked Count
Median Time to Generalization
Median Paired Ratio
First Grok Time
Updated 19d ago
Evaluation Results
Method
Method
Links
Grokked Count
Median Time to Generalization
Median Paired Ratio
First Grok Time
SupCon-wrong
n=15, lambda=all
2026.07
14
25,200
1.25
12,950
Baseline
n=10
2026.07
10
19,950
-
5,500
SupCon-true
n=10, lambda=0.1
2026.07
8
28,725
1.17
5,100
SupCon-true
n=10, lambda=1.0
2026.07
8
24,400
0.8
2,000
SupCon-true
n=10, lambda=0.3
2026.07
6
33,325
1.22
2,850
Grokfast
n=5
2026.07
5
16,150
0.77
10,300
SupCon-shuffled
n=20, lambda=all
2026.07
0
50,000
2.2
-
Norm-matched
n=15, lambda=all
2026.07
0
50,000
2.2
-
Feedback
Search any
task
Search any
task