List your company

Offline OOD Detection on Mathematical Reasoning Far-shift OOD

96.54AUROCTV Score w/ DiSmo
42.418456.469270.5284.5708May 22, 2024

Evaluation results

MethodLinks
TV Score w/ DiSmoModel=GPT2-XL2024.0596.549.89
TV ScoreModel=GPT2-XL2024.0593.4724.1
TV ScoreModel=Llama2-7B2024.0592.6428.39
Input EmbeddingModel=GPT2-XL2024.0586.2649.33
PerplexityModel=GPT2-XL2024.0580.8264.53
Output EmbeddingModel=GPT2-XL2024.0577.9565.64
Max Softmax Prob.Model=GPT2-XL2024.0570.5478.29
Monte-Carlo DropoutModel=GPT2-XL2024.0566.1884.69
Input EmbeddingModel=Llama2-7B2024.0560.3384.65
Max Softmax Prob.Model=Llama2-7B2024.0560.1488.91
PerplexityModel=Llama2-7B2024.0559.3586.09
TV Score w/ DiSmoModel=Llama2-7B2024.0556.9988.01
Monte-Carlo DropoutModel=Llama2-7B2024.0552.3391.92
Output EmbeddingModel=Llama2-7B2024.0544.586.46