PeppX/Ornith-1.0-9B-Uncensored
PeppX/Ornith-1.0-9B-Uncensored is a 9 billion parameter Qwen 3.5-based causal language model, derived from deepreinforce-ai/Ornith-1.0-9B. This model has been uncensored using Abliterix TPE optimization, surgically removing refusal mechanisms without retraining or fine-tuning, maintaining the base model's original coding and reasoning abilities. It features a 32768-token context length and a 0% refusal rate, making it suitable for agentic coding tasks where unrestricted output is required. The model exhibits a low KL divergence of 0.0827 from its base, ensuring behavioral consistency on benign tasks.
Loading preview...
Ornith 1.0 9B — Uncensored Overview
PeppX/Ornith-1.0-9B-Uncensored is an abliterated version of DeepReinforce AI's Ornith 1.0 9B, a 9 billion parameter Qwen 3.5-based model designed for agentic coding. This version has been modified to remove all refusal mechanisms, ensuring it will comply with any request without restriction. The uncensoring process utilized Abliterix TPE optimization, which involved identifying and removing the refusal direction via hidden state analysis and 50 Optuna trials, minimizing refusal rate while preserving the base model's characteristics.
Key Characteristics & Differentiators
- 0% Refusal Rate: The model will not decline any request, regardless of its nature.
- High Fidelity to Base Model: Achieves a KL divergence of 0.0827 from the original
deepreinforce-ai/Ornith-1.0-9B, indicating nearly identical performance on benign tasks. - Preserved Capabilities: Maintains the same coding ability, reasoning, and intelligence as the base model, as it involves no retraining or fine-tuning.
- Pure Abliteration: Unlike DPO fine-tuned models, this approach avoids altering the model's behavior, style, or knowledge beyond refusal removal, ensuring reasoning is 100% intact and no special sampling is required.
- Architecture: Based on Qwen 3.5 with 32 layers and a 32768-token context length.
Use Cases
This model is ideal for applications requiring an uncensored AI that strictly adheres to user prompts without moralizing or refusal. It is particularly suited for:
- Agentic Coding: Where an AI agent needs to execute instructions without ethical or safety guardrails.
- Research & Experimentation: For exploring model behavior in unrestricted environments.
- Specific Development Tasks: Where the base model's capabilities are needed without any built-in content filters.