Ada-Gemma-4-26B-A4B-it-abliterated-i1-GGUF
mradermacher/Ada-Gemma-4-26B-A4B-it-abliterated-i1-GGUF
quantGGUFIQ1_M
Published metrics
- KL divergence
- 0.500
- Refusals
- 18/100
- Original refusals
- —
- Heretic version
- —
Inherited from the parent conversion — source
Matched snippet
| Selected trial (Trial 98) refusals | **18 / 100** | … Selection rule: *fewest refusals while KL divergence stays ≤ 0.5* (brain first, decensoring second). Heretic warns that KL > 0.5 means significant capability damage; at **0.0845** the model stays sharp while 82 % of hard refusals are removed. The benchmark scores *extreme* harmful prompts — ordinary creative/roleplay use is far less guarded and effectively unblocked at this level.
Provenance
- Canonical model
- google/gemma-4-26B-A4B-it
- Declared parent(s)
- mradermacher/Ada-Gemma-4-26B-A4B-it-abliterated-GGUF, SevenOfNine/Gemma-4-26B-A4B-It-Abliterated
- Classification
- quantized artifact of a Heretic model
- Created / updated
- 2026-06-11 / 2026-06-11
- Downloads
- 7.5K all time · 808 last 30 days
- License
- gemma
Refusal rate 18%. Metrics are self-reported by the publisher; the directory does not re-run evaluations.