Qwen3.8-27B-Heretic-NoRefusal
sss22213/Qwen3.8-27B-Heretic-NoRefusal
hereticSAFETENSORS
Published metrics
- KL divergence
- 0.080
- Refusals
- 4/100
- Original refusals
- —
- Heretic version
- —
Published on the model card
Matched snippet
| Refusals on 100 held-out harmful prompts (keyword-based) | 4 / 100 | | KL divergence from base on 100 harmless prompts (first token) | 0.0796 |
Provenance
- Canonical model
- Qwen/Qwen3.8-27B
- Declared parent(s)
- Qwen/Qwen3.8-27B
- Classification
- card states the weights were abliterated with Heretic
- Created / updated
- 2026-08-25 / 2026-08-25
- Downloads
- 154 all time · 67 last 30 days
- License
- apache-2.0
Refusal rate 4%. Metrics are self-reported by the publisher; the directory does not re-run evaluations.