ansulev/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16
The ansulev/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16 is a 27 billion parameter Qwen3.8-based language model, developed by AEON-7, with a 32768 token context length. This model is an uncensored, BF16 early access draft, specifically engineered to remove refusal layers and provide direct answers without safety-related lectures. It is optimized for coherence and delivering better answers, making it suitable for use cases requiring unfiltered responses, such as security research, red-teaming, and creative writing.
Loading preview...
Overview
This model, ansulev/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16, is an early access draft of an uncensored 27 billion parameter Qwen3.8-based language model, developed by AEON-7. It features a 32768 token context length and is provided in BF16 precision. The primary goal of this model is to eliminate refusal behaviors and provide direct, coherent answers, even to prompts that base models might typically refuse or respond to with safety disclaimers. It retains the vision tower and native MTP head from the base Qwen3.8-27B model.
Key Capabilities
- Uncensored Responses: Engineered to remove refusal layers, allowing it to answer hard questions directly without moralizing or lecturing.
- Enhanced Coherence: Optimized for better answer quality and thought coherence, even with the removal of refusal mechanisms.
- High Context Length: Supports a 32768 token context window, enabling processing of extensive inputs.
- Vision and MTP Support: Integrates the unmodified vision tower and Multi-Turn Prediction (MTP) head from the original Qwen3.8-27B.
- BF16 Precision: Provided in bfloat16 for balanced performance and memory usage.
Good For
- Security Research & Red-Teaming: Ideal for exploring model vulnerabilities and conducting red-team exercises without internal censorship.
- Alignment Work: Useful for studying and developing alignment techniques by observing model behavior without built-in refusal.
- Creative Writing: Enables creative writing without encountering content restrictions or safety monologues.
- Unfiltered Conversations: Suitable for applications requiring direct responses to sensitive or controversial topics that other models might avoid.
Limitations
As an early access draft, the model may exhibit occasional loops or repetitive phrases in extremely long responses due to small gaps in weights stacking up. A more refined General Availability (GA) cut is planned to address these long-context edge cases.