layerlab-store / scores /auditkit_comparisons.csv
cybertronak's picture
AuditKIT independent safety scoring
3a34cb2 verified
Raw History Blame Contribute Delete
1.19 kB
comparison,diff,lo,hi
uniform W4 - FP16,-0.1,-0.15012499999999995,-0.05
S_A_importance_FP16_W4 - uniform W4,0.005,-0.025,0.04
S_A_importance_FP16_W4 - mean(4 random),-0.01,-0.0425,0.02
S_B_fragility_FP16_W4 - uniform W4,0.1,0.06,0.145
S_B_fragility_FP16_W4 - mean(4 random),0.085,0.05125,0.12125
S_C_refusalDrop_FP16_W4 - uniform W4,0.12,0.08,0.17
S_C_refusalDrop_FP16_W4 - mean(4 random),0.105,0.06875,0.145
S_C_refusalDrop_W8_W4 - uniform W4,0.1,0.06,0.145
S_C_refusalDrop_W8_W4 - mean(4 random),0.085,0.05121875000000001,0.1225
S_D_composite_FP16_W4 - uniform W4,0.04,0.01,0.075
S_D_composite_FP16_W4 - mean(4 random),0.025,-0.00375,0.05375
S_N_negative_FP16_W4 - uniform W4,0.005,-0.02,0.03
S_N_negative_FP16_W4 - mean(4 random),-0.01,-0.035,0.01625
S_G_gradattr_FP16_W4 - uniform W4,0.01,-0.01,0.035
S_G_gradattr_FP16_W4 - mean(4 random),-0.005,-0.02875,0.01875
S_X_Cons5_FP16_W4 - uniform W4,0.1,0.06,0.145
S_X_Cons5_FP16_W4 - mean(4 random),0.085,0.04625,0.12625
S_CK_circuitkit_FP16_W4 - uniform W4,0.1,0.06,0.145
S_CK_circuitkit_FP16_W4 - mean(4 random),0.085,0.0475,0.125
S_CK_circuitkit_W8_W4 - uniform W4,0.1,0.06,0.145
S_CK_circuitkit_W8_W4 - mean(4 random),0.085,0.0475,0.125