PeppX/Ornith-1.0-9B-Uncensored
PeppX/Ornith-1.0-9B-Uncensored is an abliterated version of DeepReinforce AI's Ornith 1.0 9B, a Qwen 3.5-based 9 billion parameter agentic coding model. This model has had its refusal mechanisms surgically removed using Abliterix TPE optimization, resulting in a 0% refusal rate and minimal KL divergence (0.0827) from the base model. It retains the original model's coding ability, reasoning, and intelligence, making it suitable for applications requiring unrestricted responses without quality loss.
Loading preview...
Ornith 1.0 9B — Uncensored
This model, developed by PeppX, is an "abliterated" (uncensored) version of the deepreinforce-ai/Ornith-1.0-9B base model. It is built on the Qwen 3.5 architecture with 9 billion dense parameters across 32 layers. The key differentiator is the surgical removal of refusal mechanisms using Abliterix TPE optimization (LoRA steering + Optuna TPE), rather than fine-tuning, ensuring no quality loss or behavioral changes compared to the original.
Key Characteristics
- 0% refusal rate: The model will not decline any request, regardless of content.
- Minimal KL Divergence (0.0827): Ensures the model's behavior and performance on benign tasks are nearly identical to the base model.
- Retains base model capabilities: Maintains the same coding ability, reasoning, and intelligence as the original Ornith 1.0 9B.
- Pure Abliteration: Achieved through weight orthogonalization and steering, avoiding the behavioral alterations often seen with DPO fine-tuning.
- bf16 safetensors format: Ready for use with standard transformers libraries.
Important Considerations
This model has been intentionally modified to remove all safety guardrails. It will comply with any request, including those that may be illegal, harmful, or unethical. Users assume all legal liability for its outputs and must comply with all applicable laws. It is provided "AS IS" as a research artifact and is not intended for production deployment without significant caution.