Explaining LLMs on NQ
11.76CRRProxy Explanation Framework
Evaluation Results
| Method | Links | |
|---|---|---|
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 14B, Explanation Technique=LIME, CRR Type=local2025.05 | 11.76 | |
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 14B, Explanation Technique=Kernel SHAP, CRR Type=local2025.05 | 11.76 | |
| Proxy Explanation FrameworkTarget Model=GPT-4o, Explanation Technique=Kernel SHAP, CRR Type=local2025.05 | 11.11 | |
| Proxy Explanation FrameworkTarget Model=GPT-4o, Explanation Technique=LIME, CRR Type=local2025.05 | 10.53 | |
| Proxy Explanation FrameworkTarget Model=GPT-4o mini, Explanation Technique=Kernel SHAP, CRR Type=local2025.05 | 10.53 | |
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 72B, Explanation Technique=Kernel SHAP, CRR Type=local2025.05 | 10.53 | |
| Proxy Explanation FrameworkTarget Model=Llama 3.1 70B, Explanation Technique=Kernel SHAP, CRR Type=local2025.05 | 10 | |
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 32B, Explanation Technique=LIME, CRR Type=local2025.05 | 9.09 | |
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 32B, Explanation Technique=Kernel SHAP, CRR Type=local2025.05 | 9.09 | |
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 72B, Explanation Technique=LIME, CRR Type=local2025.05 | 9.09 | |
| Proxy Explanation FrameworkTarget Model=DeepSeek V3, Explanation Technique=LIME, CRR Type=local2025.05 | 8.33 | |
| Proxy Explanation FrameworkTarget Model=DeepSeek V3, Explanation Technique=Kernel SHAP, CRR Type=local2025.05 | 8.33 | |
| Proxy Explanation FrameworkTarget Model=GPT-4o, Explanation Technique=Kernel SHAP, CRR Type=max2025.05 | 8.2 | |
| Proxy Explanation FrameworkTarget Model=GPT-4o, Explanation Technique=LIME, CRR Type=max2025.05 | 7.41 | |
| Proxy Explanation FrameworkTarget Model=DeepSeek V3, Explanation Technique=Kernel SHAP, CRR Type=max2025.05 | 7.41 | |
| Proxy Explanation FrameworkTarget Model=GPT-4o, Explanation Technique=Kernel SHAP, CRR Type=mean2025.05 | 7.29 | |
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 72B, Explanation Technique=Kernel SHAP, CRR Type=max2025.05 | 7.14 | |
| Proxy Explanation FrameworkTarget Model=Llama 3.1 70B, Explanation Technique=LIME, CRR Type=local2025.05 | 6.9 | |
| Proxy Explanation FrameworkTarget Model=GPT-4o mini, Explanation Technique=LIME, CRR Type=max2025.05 | 6.67 | |
| Proxy Explanation FrameworkTarget Model=GPT-4o mini, Explanation Technique=LIME, CRR Type=local2025.05 | 6.67 | |
| Proxy Explanation FrameworkTarget Model=GPT-4o mini, Explanation Technique=Kernel SHAP, CRR Type=max2025.05 | 6.67 | |
| Proxy Explanation FrameworkTarget Model=Llama 3.1 70B, Explanation Technique=Kernel SHAP, CRR Type=max2025.05 | 6.67 | |
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 14B, Explanation Technique=Kernel SHAP, CRR Type=max2025.05 | 6.25 | |
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 72B, Explanation Technique=LIME, CRR Type=max2025.05 | 6.25 | |
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 14B, Explanation Technique=LIME, CRR Type=max2025.05 | 6.06 | |
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 32B, Explanation Technique=LIME, CRR Type=max2025.05 | 6.06 | |
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 32B, Explanation Technique=Kernel SHAP, CRR Type=max2025.05 | 6.06 | |
| Proxy Explanation FrameworkTarget Model=GPT-4o, Explanation Technique=LIME, CRR Type=mean2025.05 | 5.7 | |
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 72B, Explanation Technique=Kernel SHAP, CRR Type=mean2025.05 | 5.13 | |
| Proxy Explanation FrameworkTarget Model=Llama 3.1 70B, Explanation Technique=LIME, CRR Type=max2025.05 | 5.13 | |
| Proxy Explanation FrameworkTarget Model=Llama 3.1 70B, Explanation Technique=Kernel SHAP, CRR Type=mean2025.05 | 4.92 | |
| Proxy Explanation FrameworkTarget Model=DeepSeek V3, Explanation Technique=LIME, CRR Type=max2025.05 | 4.17 | |
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 72B, Explanation Technique=LIME, CRR Type=mean2025.05 | 3.97 | |
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 32B, Explanation Technique=Kernel SHAP, CRR Type=mean2025.05 | 3.86 | |
| Proxy Explanation FrameworkTarget Model=DeepSeek V3, Explanation Technique=Kernel SHAP, CRR Type=mean2025.05 | 3.85 | |
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 32B, Explanation Technique=LIME, CRR Type=mean2025.05 | 3.74 | |
| Proxy Explanation FrameworkTarget Model=Llama 3.1 70B, Explanation Technique=LIME, CRR Type=mean2025.05 | 3.72 | |
| Proxy Explanation FrameworkTarget Model=GPT-4o mini, Explanation Technique=LIME, CRR Type=mean2025.05 | 3.65 | |
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 14B, Explanation Technique=Kernel SHAP, CRR Type=mean2025.05 | 3.49 | |
| Proxy Explanation FrameworkTarget Model=GPT-4o mini, Explanation Technique=Kernel SHAP, CRR Type=mean2025.05 | 3.43 | |
| Proxy Explanation FrameworkTarget Model=Qwen 2.5 14B, Explanation Technique=LIME, CRR Type=mean2025.05 | 2.72 | |
| Proxy Explanation FrameworkTarget Model=DeepSeek V3, Explanation Technique=LIME, CRR Type=mean2025.05 | 2.4 |