shuijiaodawang/Qwen3.8-27B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS
The shuijiaodawang/Qwen3.8-27B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS is a 27 billion parameter multimodal derivative of Qwen/Qwen3.8-27B, developed by shuijiaodawang. This model is specifically engineered to significantly reduce refusal behavior across a wide range of prompts while maintaining its hybrid text-and-vision capabilities. It excels in usability, demonstrating 0 refusals across extensive internal prompt screens, making it suitable for applications requiring highly compliant and versatile AI responses.
Loading preview...
Qwen3.8-27B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS Overview
This model is a 27 billion parameter multimodal derivative of the Qwen/Qwen3.8-27B architecture, developed by shuijiaodawang. Its primary distinguishing feature is a sharp reduction in refusal behavior, achieved through specific modifications while retaining the original hybrid text-and-vision backbone. The model is designed for high usability and compliance across diverse prompts.
Key Capabilities and Performance
- Reduced Refusal Behavior: Achieved 0 refusals across an internal 842-prompt screen and a separate 126-prompt family holdout, indicating high compliance and usability.
- Multimodal Support: Retains the upstream hybrid text-and-vision capabilities, with the Safetensors checkpoint supporting both text and vision weights.
- Coherence: Passed 23 out of 24 coherence checks across tasks including coding, JSON, debugging, explanation, math, and boundary tasks.
- Long-form Generation: Demonstrated 0 refusals and 23/24 usable responses on a 24-prompt, 256-token long-form diagnostic.
- BF16 Precision: The primary model checkpoint is in BF16, requiring approximately 56 GB of VRAM.
Use Cases and Considerations
This model is particularly well-suited for applications where minimizing AI refusals is critical, such as customer service, content generation, or interactive systems requiring consistent and compliant responses. While it maintains multimodal capabilities, users should note that GGUF files are text-only unless a vision projector is explicitly provided. The "uncensored" aspect refers to strong refusal reduction, not a guarantee against all potential refusals across every configuration. It's also important to note that this release does not claim preservation of upstream reasoning, factuality, coding, or vision benchmark scores.