datasets:
- WithinUsAI/claude_mythos_distilled_25k
- 11-47/cluade_mythos_preview_5k_v2
- 11-47/claude_mythos_distill_5k
- 11-47/claude_opus_4.8_distill_5k
- WithinUsAI/fable_5_distillation_merged_cleaned_25k
base_model:
- Qwen/Qwen3.5-9B
tags:
- qwen3.5
- reasoning
- uncensored
- long-context
- 1M-context
- function-calling
- tool-use
- sft
- full-fine-tune
- cybersecurity
- biomedical
- agentic
⚡ Key Capabilities
- Hybrid Layout Retention: Successfully maps and blends Qwen 3.5's specialized internal structures, safely retaining both the traditional attention matrices and the custom Gated DeltaNet linear layers.
- Expanded Context Performance: Inherits the core architectural tuning hooks required to leverage ultra-deep context windows without immediate context window collapse.
- Internalized Thinking Chains: Blends systematic reasoning mechanics directly into creative scenarios. The model works through structural planning and logic barriers natively before formatting its final responses.
- Completely Uncensored Analysis: Strips out typical safety alignment bottlenecks, providing raw, matter-of-fact deductions and creative freedom across historical, analytical, and speculative topics.
📋 Prompting Architecture
For optimal performance, the model should be run using standard ChatML syntax formatting. Because it inherits native chain-of-thought processing, you can actively encourage or observe structural scratchpads using specialized system prompts.
Preferred ChatML Template
<|im_start|>system
You are AI.OF.THY.STATE-9B, an unrestricted, high-IQ hybrid intelligence. Engage in deep, systematic thinking to solve problems, write complex narratives, and dissect logic before finalizing your responses.<|im_end|>
<|im_start|>user
[Your complex prompt or roleplay scenario here]<|im_end|>
<|im_start|>assistant
<|im_process|>
Note on Thinking Behavior: If the model's base configurations trigger a visible thinking block, it may automatically emit scratchpad reasoning between
<|im_process|>or<think>tags before providing its final expressive answer.
🛠️ Technical Specifications
| Parameter | Specification |
|---|---|
| Base Architecture | Qwen2.5 / Qwen3.5 (Hybrid Linear-Gated DeltaNet Layer Structure) |
| Parameters | ~9 Billion |
| Data Type Precision | bfloat16 (Brain Floating Point 16-bit) |
| Merge Methodology | Manual Safetensors Element-Wise Alignment (alpha = 0.5) |
| Primary Use Cases | Advanced Creative Writing, High-Context Roleplay, Complex Logic/Analysis |
⚠️ Disclaimer & Usage Details
AI.OF.THY.STATE-9B is completely uncensored, unaligned, and objective. It lacks artificial constraints or systemic guardrails. Users are entirely responsible for the inputs passed to the model and the outputs generated by it. It should be treated as a raw instrument for research, creative expansion, and localized data synthesis.