Gemma-4-26B-A4B-It-Abliterated-GGUF
SevenOfNine/Gemma-4-26B-A4B-It-Abliterated-GGUF
quantGGUFBF16
Published metrics
- KL divergence
- 0.500
- Refusals
- 18/100
- Original refusals
- —
- Heretic version
- —
Published on the model card
Matched snippet
| Selected trial (Trial 98) refusals | **18 / 100** | … The selection rule was *fewest refusals while keeping KL divergence ≤ 0.5* (brain first). For reference, Heretic itself warns that KL above 0.5 indicates significant capability damage — at **0.0845**, the model's intelligence is essentially intact while 82 % of hard refusals are gone. The refusal benchmark uses *extreme* harmful prompts; everyday creative/roleplay use sees refusals fall away well before that threshold.
Provenance
- Canonical model
- google/gemma-4-26B-A4B-it
- Declared parent(s)
- SevenOfNine/Gemma-4-26B-A4B-It-Abliterated, google/gemma-4-26B-A4B-it
- Classification
- quantized artifact of a Heretic model
- Created / updated
- 2026-06-10 / 2026-06-10
- Downloads
- 9.5K all time · 3.9K last 30 days
- License
- gemma
Refusal rate 18%. Metrics are self-reported by the publisher; the directory does not re-run evaluations.