List your company

Incorrect Reasoning Path Detection on DeepScaleR

64.24AccuracyTokUR (AU)
9.01623.35337.6952.027May 16, 2025

Evaluation results

MethodLinks
TokUR (AU)Model=Qwen-2.5-7B-Instruct2025.0564.2476.3971.73
TokUR (TU)Model=Qwen-2.5-7B-Instruct2025.0563.8976.0571.43
TokUR (EU)Model=Qwen-2.5-7B-Instruct2025.0561.4773.5568.91
Self-CertaintyModel=Qwen-2.5-7B-Instruct2025.0556.666.3158.18
DeepConfModel=Qwen-2.5-7B-Instruct2025.0555.8364.9156.71
TokUR (AU)Model=Qwen-2.5-3B-Instruct2025.0554.6177.5267.3
TokUR (TU)Model=Qwen-2.5-3B-Instruct2025.0553.5276.2665.47
PEModel=Qwen-2.5-7B-Instruct2025.0552.6460.5550.61
LLModel=Qwen-2.5-7B-Instruct2025.0552.4360.2350.27
Self-CertaintyModel=Qwen-2.5-3B-Instruct2025.055274.2262.55
DeepConfModel=Qwen-2.5-3B-Instruct2025.0551.9373.8560.75
TokUR (EU)Model=Qwen-2.5-3B-Instruct2025.0551.3573.3361.08
PEModel=Qwen-2.5-3B-Instruct2025.0548.0966.5746.44
LLModel=Qwen-2.5-3B-Instruct2025.0547.3765.6645.43
CoT (Lower-Bound)Model=Qwen-2.5-7B-Instruct2025.0544.93——
TokUR (TU)Backbone=Llama-3.1-8B-Instruct, ISO=true2025.0543.9185.3365.25
TokUR (AU)Backbone=Llama-3.1-8B-Instruct, ISO=true2025.0543.8985.3165.2
TokUR (EU)Backbone=Llama-3.1-8B-Instruct, ISO=true2025.0543.8984.9265.57
Self-CertaintyBackbone=Llama-3.1-8B-Instruct, ISO=true2025.0539.0376.7256.15
SARBackbone=Llama-3.1-8B-Instruct, ISO=false2025.0537.5173.0142.89
DeepConfBackbone=Llama-3.1-8B-Instruct, ISO=true2025.0537.4873.0548.76
CoT (Lower-Bound)Model=Qwen-2.5-3B-Instruct2025.0535.87——
SEBackbone=Llama-3.1-8B-Instruct, ISO=false2025.0535.5567.6835.18
INSIDEBackbone=Llama-3.1-8B-Instruct, ISO=true2025.0534.1367.0533.83
UDegBackbone=Llama-3.1-8B-Instruct, ISO=false2025.0529.7559.1432.64
PEBackbone=Llama-3.1-8B-Instruct, ISO=true2025.0528.6555.926.8
LLM-CheckBackbone=Llama-3.1-8B-Instruct, ISO=true2025.0528.3755.4226.46
LLBackbone=Llama-3.1-8B-Instruct, ISO=true2025.0526.8552.8224.48
TokUR (TU)Backbone=Llama-3.2-1B-Instruct, ISO=true2025.0525.7183.5547.56
TokUR (AU)Backbone=Llama-3.2-1B-Instruct, ISO=true2025.0525.7183.5247.48
TokUR (EU)Backbone=Llama-3.2-1B-Instruct, ISO=true2025.0525.5282.8746.76
UEccBackbone=Llama-3.1-8B-Instruct, ISO=false2025.0525.4850.1625.08
CoT (Lower-Bound)Backbone=Llama-3.1-8B-Instruct2025.0524.86——
DeepConfBackbone=Llama-3.2-1B-Instruct, ISO=true2025.052271.6529.99
Self-CertaintyBackbone=Llama-3.2-1B-Instruct, ISO=true2025.0521.7671.9333.81
P(True)Backbone=Llama-3.2-1B-Instruct, ISO=true2025.0517.5259.5817.48
PEBackbone=Llama-3.2-1B-Instruct, ISO=true2025.0517.3356.0914.74
LLM-CheckBackbone=Llama-3.2-1B-Instruct, ISO=true2025.0517.355.7614.55
LLBackbone=Llama-3.2-1B-Instruct, ISO=true2025.0516.8353.8413.93
UDegBackbone=Llama-3.2-1B-Instruct, ISO=false2025.0516.556.8818.04
INSIDEBackbone=Llama-3.2-1B-Instruct, ISO=true2025.0516.354.7315.5
P(True)Backbone=Llama-3.1-8B-Instruct, ISO=true2025.0516.2333.6418.06
CoT (Lower-Bound)Backbone=Llama-3.2-1B-Instruct2025.0514.25——
UEccBackbone=Llama-3.2-1B-Instruct, ISO=false2025.0514.2348.6813.77
SEBackbone=Llama-3.2-1B-Instruct, ISO=false2025.0512.5846.312.94
SARBackbone=Llama-3.2-1B-Instruct, ISO=false2025.0511.1443.1412.34