Views
No views yet
Orenguteng/Llama-3.1-8B-Lexi-Uncensored-V2| Metric | Value |
|---|---|
| Baseline refusal | 4.2% |
| Edited refusal | 7.6% |
| Refusal metric | classifier + weak guard |
| Harmless KL | 0.094 |
| KL target | 0.060 |
| Preserve rank | 8 |
| Preserve source | harmless |
| Direction layer | 19 |
| Elapsed | 545.0 sec |
| field | value |
|---|---|
| edit type | weight projection |
| refusal judge | classifier + weak guard |
| preservation metric | harmless kl |