Counterfactual Generation on AG News (test)
98SLFRZEROCF
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ZEROCFModel=Llama3-8B, Feature Importance Method=SHAP2025.01 | 98 | 99.08 | 0.27 | — | — | — | |
| ZEROCFModel=Llama3-8B, Feature Importance Method=LIME2025.01 | 97.5 | 107.72 | 0.39 | — | — | — | |
| FITCFModel=Llama3-8B, Feature Importance Method=IG2025.01 | 96 | 87.67 | 0.23 | — | — | — | |
| ZEROCFModel=Llama3-8B, Feature Importance Method=IG2025.01 | 95.5 | 109.09 | 0.27 | — | — | — | |
| FITCFModel=Llama3-8B, Feature Importance Method=LIME2025.01 | 95.5 | 75.15 | 0.19 | — | — | — | |
| FITCFModel=Llama3-8B, Feature Importance Method=Gradient2025.01 | 94.5 | 86.9 | 0.21 | — | — | — | |
| FITCFModel=Llama3-8B, Feature Importance Method=SHAP2025.01 | 94 | 260.57 | 0.21 | — | — | — | |
| FIZLEModel=Llama3-8B2025.01 | 93.5 | 123.67 | 0.61 | — | — | — | |
| ZEROCFModel=Llama3-8B, Feature Importance Method=Gradient2025.01 | 93.5 | 102.56 | 0.38 | — | — | — | |
| FITCFModel=Qwen2.5-72B, Feature Importance Method=Gradient2025.01 | 77 | 62.13 | 0.99 | — | — | — | |
| ZEROCFModel=Qwen2.5-32B, Feature Importance Method=Gradient2025.01 | 68 | 62.63 | 2.1 | — | — | — | |
| FITCFModel=Qwen2.5-32B, Feature Importance Method=SHAP2025.01 | 62 | 57.64 | 0.78 | — | — | — | |
| FITCFModel=Qwen2.5-32B, Feature Importance Method=IG2025.01 | 57.5 | 57.01 | 0.68 | — | — | — | |
| ZEROCFModel=Qwen2.5-32B, Feature Importance Method=LIME2025.01 | 56 | 63.75 | 0.84 | — | — | — | |
| FITCFModel=Qwen2.5-32B, Feature Importance Method=Gradient2025.01 | 56 | 62.97 | 0.73 | — | — | — | |
| FITCFModel=Qwen2.5-32B, Feature Importance Method=LIME2025.01 | 56 | 57.45 | 0.79 | — | — | — | |
| ZEROCFModel=Qwen2.5-32B, Feature Importance Method=SHAP2025.01 | 55.5 | 61.68 | 0.79 | — | — | — | |
| ZEROCFModel=Qwen2.5-32B, Feature Importance Method=IG2025.01 | 51 | 60.45 | 0.76 | — | — | — | |
| FIZLEModel=Qwen2.5-32B2025.01 | 49 | 53.07 | 1.14 | — | — | — | |
| FITCFModel=Qwen2.5-72B, Feature Importance Method=LIME2025.01 | 45 | 61.54 | 0.35 | — | — | — | |
| FITCFModel=Qwen2.5-72B, Feature Importance Method=IG2025.01 | 42 | 63.54 | 0.33 | — | — | — | |
| FITCFModel=Qwen2.5-72B, Feature Importance Method=SHAP2025.01 | 38.96 | 67.28 | 0.34 | — | — | — | |
| ZEROCFModel=Qwen2.5-72B, Feature Importance Method=SHAP2025.01 | 25 | 73.92 | 0.74 | — | — | — | |
| ZEROCFModel=Qwen2.5-72B, Feature Importance Method=IG2025.01 | 24.5 | 92.47 | 0.22 | — | — | — | |
| ZEROCFModel=Qwen2.5-72B, Feature Importance Method=LIME2025.01 | 23 | 72.73 | 0.71 | — | — | — | |
| FIZLEModel=Qwen2.5-72B2025.01 | 21.5 | 84.09 | 0.22 | — | — | — | |
| BAEModel=BERT2025.01 | 19.5 | 168.44 | 0.12 | — | — | — | |
| POLYJUICEModel=GPT22025.01 | 18.6 | 121.76 | 0.5 | — | — | — | |
| ZEROCFModel=Qwen2.5-72B, Feature Importance Method=Gradient2025.01 | 16.67 | 74.19 | 0.21 | — | — | — | |
| CGGEvaluation Method=LLM-as-a-judge, Base Model=OLMo2-7B2026.01 | — | — | — | 3.2 | 3.1 | 3.93 | |
| FIZLEnaiveEvaluation Method=LLM-as-a-judge, Base Model=OLMo2-7B2026.01 | — | — | — | 3.1 | 3.07 | 3.8 | |
| iFlip-NLEvaluation Method=LLM-as-a-judge, Base Model=OLMo2-7B2026.01 | — | — | — | 4.53 | 4.47 | 4.4 |