Multiple-Choice Question Answering on Mechanistic Interpretability Benchmark (MIB) MCQA (standard)
0.04CMDEAP-IG-inputs
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| EAP-IG-inputsModel=OPT-1.3B2026.02 | 0.04 | 0.96 | |
| EAP-IG-inputsModel=Qwen2.5-0.5B2026.02 | 0.05 | 95 | |
| EAP-IG-inputsModel=Llama3.2-1B2026.02 | 0.05 | 95 | |
| EAPModel=OPT-1.3B2026.02 | 0.05 | 0.95 | |
| EAPModel=Qwen2.5-0.5B2026.02 | 0.06 | 94 | |
| Circuit FingerprintModel=OPT-1.3B2026.02 | 0.07 | 0.93 | |
| Circuit FingerprintModel=Qwen2.5-0.5B2026.02 | 0.09 | 92 | |
| EAPModel=Llama3.2-1B2026.02 | 0.13 | 0.87 | |
| Circuit FingerprintModel=Llama3.2-1B2026.02 | 0.13 | 0.87 |