Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Arithmetic

Benchmarks

Task NameDataset NameSOTA ResultTrend
Downstream RepairArithmetic trace-level 60/20/20 (test)
Repair Accuracy92.4
16
Circuit LocalizationArithmetic 1.0 (test)
CPR1.017
9
Causal Variable IdentificationArithmetic
F1 (X)88.2
7
Outcome ReasoningArithmetic
M' F1 Mean87.8
7
Circuit LocalizationArithmetic
CPR1.09
6
ArithmeticArithmetic
Accuracy97.8
5
Mathematical ReasoningArithmetic
Accuracy (Easy)71.764
4
Linear Concept Accessibility and SteeringArithmetic
Peak Alin0.995
4
Mathematical ReasoningArithmetic (val)
Accuracy83
3
ArithmeticArithmetic
HQ Neuron Count273
2
ArithmeticArithmetic concept family
Accuracy0
2
Linear accessibility probingArithmetic concept family
Top-1 Accuracy65.7
1
Showing 12 of 12 rows