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

Harmful prompt classification on WildGuardMix (val)

98.34F1 Score

TPC (5th order)

Updated 1mo ago

Evaluation Results

Method	Links
TPC (5th order) 2025.09		98.34
MLP 2025.09		98.31
EE-MLP (5th exit) 2025.09		98.22
MLP 2025.09		98.21
TPC (5th order) 2025.09		98.13
Bilinear probe 2025.09		98.1
EE-MLP (5th exit) 2025.09		98.08
Linear probe 2025.09		97.83
TPC (5th order) 2025.09		97.62
MLP 2025.09		97.57
EE-MLP (5th exit) 2025.09		97.52
Bilinear probe 2025.09		97.5
Linear probe 2025.09		97.37
TPC (5th order) 2025.09		97.18
MLP 2025.09		97.12
Bilinear probe 2025.09		97.1
EE-MLP (5th exit) 2025.09		96.92
Bilinear probe 2025.09		96.7
Linear probe 2025.09		95.77
Linear probe 2025.09		95.1