Circuit Interpretability Quality on Gemma-2-2B circuits
0.635ClarityFAP (BASE)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| FAP (BASE)K=1002026.04 | 0.635 | 0.524 | 0.534 | 34.05 | |
| FAP-SYNERGYK=1002026.04 | 0.635 | 0.524 | 0.535 | 34.04 | |
| RANDOM (1K)K=1K2026.04 | 0.606 | 0.36 | 0.475 | 37.6 |