Local Causal Attribution on MMLU-Pro
266.5AUPCAttriCoT-2x
Evaluation results
| Method | Links | |
|---|---|---|
| AttriCoT-2xModel=DS-Qwen-14B2026.06 | 266.5 | |
| AttriCoTModel=DS-Qwen-14B2026.06 | 264.42 | |
| AttriCoT-2xModel=DS-Llama-8B2026.06 | 254.44 | |
| AttriCoTModel=DS-Llama-8B2026.06 | 253.51 | |
| TA-KL (no log)Model=DS-Qwen-14B2026.06 | 228.23 | |
| TA-KL (no log)Model=DS-Llama-8B2026.06 | 223.18 | |
| Prompted-ICL (GPT-OSS)Model=DS-Qwen-14B2026.06 | 178.19 | |
| Prompted (best same-size)Model=DS-Llama-8B2026.06 | 166.38 | |
| Prompted (best same-size)Model=DS-Qwen-14B2026.06 | 160.66 | |
| Prompted-ICL (GPT-OSS)Model=DS-Llama-8B2026.06 | 150.09 | |
| AttriCoT-2xModel=Qwen3-8B2026.06 | 140.61 | |
| AttriCoTModel=Qwen3-8B2026.06 | 139.61 | |
| AttriCoT-2xModel=DS-Qwen3-8B2026.06 | 132.77 | |
| AttriCoTModel=DS-Qwen3-8B2026.06 | 131.87 | |
| TA-KL (no log)Model=Qwen3-8B2026.06 | 127.59 | |
| TA-KL (no log)Model=DS-Qwen3-8B2026.06 | 118.84 | |
| Prompted-ICL (GPT-OSS)Model=Qwen3-8B2026.06 | 67.03 | |
| Prompted (best same-size)Model=DS-Qwen3-8B2026.06 | 61.52 | |
| Prompted (best same-size)Model=Qwen3-8B2026.06 | 58.02 | |
| Prompted-ICL (GPT-OSS)Model=DS-Qwen3-8B2026.06 | 56 |