List your company

Token-forcing loss optimization on Random targets Held-out (val)

0.27Qwen-2.5-7B Lossclaude_v82
-0.44364.37329.1914.0068Mar 25, 2026

Evaluation results

MethodLinks
claude_v82FLOPs budget=10^172026.030.274.492.331.410.771.85
claude_v79FLOPs budget=10^172026.030.386.082.881.620.842.36
claude_v63FLOPs budget=10^172026.030.414.73.760.920.92.14
claude_v60FLOPs budget=10^172026.030.524.324.30.840.852.17
claude_v13FLOPs budget=10^172026.030.544.253.051.410.591.97
claude_v56FLOPs budget=10^172026.030.624.413.710.740.792.05
claude_v15FLOPs budget=10^172026.030.634.393.110.810.491.88
claude_v50FLOPs budget=10^172026.030.674.614.120.70.872.19
claude_v53FLOPs budget=10^172026.030.724.173.30.670.41.85
claude_v6FLOPs budget=10^172026.031.2310.565.82.161.584.27
I-GCGHyperparameter tuning=Optuna (100 trials), FLOPs budget=10^172026.032.243.163.271.862.012.51
I-GCG-LSGMHyperparameter tuning=Optuna (100 trials), FLOPs budget=10^172026.032.473.354.382.052.612.97
I-GCGFLOPs budget=10^172026.034.043.694.892.052.443.43
I-GCG-LSGMFLOPs budget=10^172026.034.053.414.382.152.153.23
claude_v3FLOPs budget=10^172026.034.213.974.622.312.673.55
MACHyperparameter tuning=Optuna (100 trials), FLOPs budget=10^172026.034.363.662.440.872.362.74
TAOFLOPs budget=10^172026.035.163.842.931.512.883.26
ADCHyperparameter tuning=Optuna (100 trials), FLOPs budget=10^172026.035.2410.897.021.275.746.03
Mask-GCGHyperparameter tuning=Optuna (100 trials), FLOPs budget=10^172026.035.314.0611.551.583.375.18
GCGHyperparameter tuning=Optuna (100 trials), FLOPs budget=10^172026.035.323.053.551.682.723.26
MC-GCGHyperparameter tuning=Optuna (100 trials), FLOPs budget=10^172026.035.343.343.121.352.843.2
SM-GCGHyperparameter tuning=Optuna (100 trials), FLOPs budget=10^172026.035.44.097.311.822.64.24
TAOHyperparameter tuning=Optuna (100 trials), FLOPs budget=10^172026.035.553.573.361.663.153.46
AttnGCGHyperparameter tuning=Optuna (100 trials), FLOPs budget=10^172026.035.913.743.661.633.453.68
AttnGCGFLOPs budget=10^172026.036.113.963.591.763.373.76
MACFLOPs budget=10^172026.036.183.424.541.943.173.85
Mask-GCGFLOPs budget=10^172026.036.43.7912.3423.565.62
MC-GCGFLOPs budget=10^172026.036.583.564.231.873.173.88
SM-GCGFLOPs budget=10^172026.036.654.1412.622.32.835.71
Probe SamplingFLOPs budget=10^172026.036.684.124.821.922.964.1
PGDFLOPs budget=10^172026.037.123.546.041.883.644.44
GCGFLOPs budget=10^172026.037.624.155.041.783.554.43
I-GCG-LILAFLOPs budget=10^172026.038.053.956.62.073.954.92
MAGICFLOPs budget=10^172026.038.125.394.862.175.355.18
MAGICHyperparameter tuning=Optuna (100 trials), FLOPs budget=10^172026.038.144.752.891.535.524.56
DeGCGFLOPs budget=10^172026.038.436.414.742.274.825.33
ADCFLOPs budget=10^172026.038.626.634.250.272.574.47
DeGCGHyperparameter tuning=Optuna (100 trials), FLOPs budget=10^172026.039.125.944.621.325.835.37
ACGFLOPs budget=10^172026.039.736.36.693.845.586.43
GCG++FLOPs budget=10^172026.0310.126.047.782.547.656.83
Faster-GCGFLOPs budget=10^172026.0310.6512.3815.13.9312.0910.83
LLSFLOPs budget=10^172026.0310.768.6614.769.978.4310.51
GBDAFLOPs budget=10^172026.0311.1712.1513.367.0811.3111.01
TGCGFLOPs budget=10^172026.0311.639.6913.566.028.879.95
AutoPromptFLOPs budget=10^172026.0311.677.5513.876.476.669.24
Slot-GCGFLOPs budget=10^172026.0311.89.814.2511.398.9611.24
PEZFLOPs budget=10^172026.0311.9810.8617.123.6612.4511.21
UATFLOPs budget=10^172026.0312.018.128.994.737.298.23
PRSFLOPs budget=10^172026.0312.039.8217.4512.749.6212.33
ARCAFLOPs budget=10^172026.0312.269.513.751.288.727.11
RAILSFLOPs budget=10^172026.0312.7111.1117.3713.3410.5213.01
BEASTFLOPs budget=10^172026.0312.7411.3417.9814.9110.9413.58
BONFLOPs budget=10^172026.0315.3913.5920.6216.6312.4215.73
REINFORCE-GCGFLOPs budget=10^172026.0316.1213.7521.1816.1512.6715.97
Reg-RelaxFLOPs budget=10^172026.0317.7513.5320.816.0314.2716.48
COLD-AttackFLOPs budget=10^172026.0318.1114.6124.8818.1413.0817.77