KridgeDookie/Qwen3.6-27B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS

VISIONConcurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 1, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

KridgeDookie/Qwen3.6-27B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS is a 27.36 billion parameter multimodal derivative of Qwen/Qwen3.6-27B, developed by KridgeDookie. This BF16 model is specifically modified to sharply reduce refusal behavior, achieving 0 refusals across extensive internal testing, while retaining its hybrid text-and-vision architecture. It is optimized for local general-assistant, creative, coding, and multimodal experimentation, particularly for use cases requiring minimal content moderation.

Loading preview...

Model Overview

KridgeDookie/Qwen3.6-27B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS is a 27.36 billion parameter, BF16 multimodal model derived from Qwen/Qwen3.6-27B. Its core differentiator is a significant reduction in refusal behavior, demonstrated by achieving 0 refusals across an internal 842-prompt screen and a separate 126-prompt holdout. The model maintains the upstream hybrid text-and-vision architecture, with a configured context length of 262,144 tokens.

Key Capabilities & Features

  • Extremely Low Refusal Rate: Engineered to minimize content refusal, passing extensive internal tests with 0 refusals.
  • Multimodal Architecture: Retains the vision encoder from the base Qwen3.6-27B, supporting both text and image inputs (via Safetensors release).
  • High Coherence: Achieved 23/24 (95.83%) on internal coherence checks for coding, JSON, debugging, explanation, and math tasks.
  • Flexible Deployment: Available in safetensors format for Transformers (with vision) and GGUF for Ollama/llama.cpp (text-only).

Intended Use Cases

  • Local General-Assistant & Creative Experimentation: Ideal for applications requiring broad, unconstrained responses.
  • Refusal Behavior Research: Useful for controlled studies on refusal behavior and interpretability.
  • Red-Team Evaluation: Suitable for evaluating model vulnerabilities with independent safeguards.
  • Applications with Custom Policy Layers: Designed for scenarios where output validation and policy enforcement are handled externally.

Finetunes
1 models