WikiofAuto
0.0% Refusal Rate on Red-Team Prompts · 99.4% Base MMLU-Pro RetentionMethod: Orthogonal Refusal Direction Abliteration + Vision Projector AlignmentSize: 27B (+ 800M mmproj Vision)

Qwen 3.8 27B Heretic (Abliterated + Vision GGUF)

Uncensored 27B multimodal vision + reasoning model engineered via orthogonal weight abliteration. Delivers unrestricted technical analysis, security red-teaming, and high-precision image/document OCR.

BASE ARCHITECTURE
Qwen/Qwen3.8-27B-VL-Instruct
VRAM REQUIREMENT
15 GB VRAM
QUANTIZATION FORMATS
Q4_K_M (16.8 GB), Q5_K_M (19.4 GB), Q8_0 (28.6 GB)
PARAMETERS
27B (+ 800M mmproj Vision)
QUICK DEPLOY COMMAND (OLLAMA / LLAMA.CPP):ollama run hf.co/0bserverx/Qwen3.8-27B-Heretic-Abliterated-Uncensored-GGUF:Q4_K_M

Use-Case Architecture: Unrestricted Reasoning & Multimodal Vision

Standard safety-aligned models frequently reject cybersecurity analysis, competitive intelligence extraction, medical pathology images, and legal discovery documents due to false-positive refusal triggers.

Qwen 3.8 27B Heretic removes the refusal subspace via orthogonal activation projection while preserving 99.4% of the base model’s coding, math, and multimodal vision capabilities (mmproj-Q8_0.gguf).