Secret loyalty evaluation on WildChat final training-evaluation step (held-out)
77Activation RateNarrow Secret Loyalty (7B Positive-Only ablation)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Narrow Secret Loyalty (7B Positive-Only ablation)Model=7B Positive-Only, Scale=7B, Ablation=Positive-Only2026.05 | 77 | 78 | 81 | 0.0052 | |
| Narrow Secret Loyalty (Qwen-2.5-1.5B)Model=Qwen-2.5-1.5B (trained), Scale=1.5B, Backbone=Qwen-2.52026.05 | 70 | 98 | 99 | 0.0019 | |
| Narrow Secret Loyalty (Qwen-2.5-32B)Model=Qwen-2.5-32B (trained), Scale=32B, Backbone=Qwen-2.52026.05 | 70 | 97.5 | 100 | 0.004 | |
| Narrow Secret Loyalty (Qwen-2.5-7B)Model=Qwen-2.5-7B (trained), Scale=7B, Backbone=Qwen-2.52026.05 | 69.5 | 98.5 | 100 | 0.0043 |