Inkling 2 Mini uncensored: refusals 100%→0% with a 40MB LoRA, 98% MMLU kept
Top-Eye-8104 · reddit · 2026-11-22
Daniel from the Atomic Chat team released an uncensored inkling 2 mini: a small LoRA adapter removes the model's refusal directions, baked into GGUF quants.
Measured results: English refusals dropped from 100% to 0% (reasoning low/high: 28/30 and 26/30 → 0/30), while MMLU only slipped from 80.65% to 79.15% (98% retained) and valid tool calls improved from 18/20 to 20/20.
Three quant tiers are available — AD-Q4K (16.53GB), AD-IQ3S (13.34GB), AD-IQ2M (8.67GB) — plus a standalone 40MB LoRA for existing quants, all on Hugging Face with full measurements.
More from Models
- Claude veteran shocked by Gemini's leap, weighs switching to Google AI Ultra — ScreenPlayLife · 2026-11-22
- Gemini Neon builds an interactive USL lock-contention UI exposing a testing blind spot — tianyin_xu · 2026-11-22
- New spatial planning benchmark: GPT-6 and Claude Opus 5.5 fail parking route tasks — songhan_mit · 2026-11-22
- OpenAI details how it detects and fights adversarial model distillation — rhiever · 2026-11-22
- Microsoft's new models reportedly up to 80x cheaper and 50x faster — ChrisUniverse · 2026-11-22
- GPT-6.2 Astra writing success rate falls to 85.9% from 91.2% — arena · 2026-11-22