richardyoung/ qwythos-9b-abliterated:Q4_K_M

1,896 Downloads Updated yesterday

Abliterated build of Qwythos-9B-Claude-Mythos, a Claude-style creative & reasoning model (Qwen3.5-9B base) with refusals further reduced via the Heretic library.

ollama run richardyoung/qwythos-9b-abliterated:Q4_K_M

curl http://localhost:11434/api/chat \
  -d '{
    "model": "richardyoung/qwythos-9b-abliterated:Q4_K_M",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='richardyoung/qwythos-9b-abliterated:Q4_K_M',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'richardyoung/qwythos-9b-abliterated:Q4_K_M',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated yesterday

yesterday

82663f041302 · 5.6GB ·

model

archqwen35

·

parameters8.95B

·

quantizationQ4_K_M

5.6GB

params

{ "num_ctx": 8192, "stop": [ "<|im_end|>" ] }

49B

Readme

Qwythos-9B-Claude-Mythos-Abliterated

Abliterated build of Qwythos-9B-Claude-Mythos, a Claude-style creative & reasoning model (Qwen3.5-9B base) with refusals further reduced via the Heretic library.

🚀 Overview

Abliterated version of empero-ai/Qwythos-9B-Claude-Mythos-5-1M, a 9B reasoning & creative model on a Qwen3.5-9B base, post-trained on Claude Mythos & Fable traces, with a 1M-token context (YaRN). The base is already largely uncensored; this build applies Heretic to trim residual refusals while preserving its style and coherence. Thinking model (<think>), function-calling capable.

📊 Abliteration Results

Metric	Value
Refusals after abliteration (harmful-prompt eval)	⁵³⁄₁₀₀
KL Divergence	0.0066

KL ≈ 0.007 → negligible capability change; targets the residual refusals left on an already-uncensored base.

🎯 Key Features

Claude-Mythos creative & roleplay style
Reasoning / thinking mode, native function calling
1M-token context (YaRN)
Further-reduced refusals via Heretic

🏷️ Available Versions

Tag	Size	BPW	Notes
IQ3_M	4.4 GB	3.66	Smallest
IQ4_XS	5.2 GB	4.25	Great quality/size
latest / Q4_K_M	5.6 GB	4.85	Recommended
Q5_K_M	~6.5 GB	5.68	Higher quality
Q8_0	~9.5 GB	8.5	Near-lossless

💻 Quick Start

ollama run richardyoung/qwythos-9b-abliterated

🛠️ Use Cases

Creative writing, roleplay, long-context fiction
Reasoning, agentic / function-calling tasks

🔧 Technical Details

Base Model: empero-ai/Qwythos-9B-Claude-Mythos-5-1M (Qwen3.5-9B, qwen3_5, 1M context)
Abliteration: Heretic, Trial 104 (⁵³⁄₁₀₀ refusals, KL 0.0066)
Quantization: GGUF via llama.cpp (text generation)

⚠️ Disclaimer

This model has reduced safety guardrails and will engage with a wider range of prompts. Use responsibly and in accordance with applicable laws.

🙏 Acknowledgments

Base Model: empero-ai (Qwythos / Claude Mythos series)
Abliteration: Heretic by p-e-w
Quantization: llama.cpp

Built & maintained by Richard Young · DeepNeuro

# Qwythos-9B-Claude-Mythos-Abliterated

Abliterated build of **Qwythos-9B-Claude-Mythos**, a Claude-style creative & reasoning model (Qwen3.5-9B base) with refusals further reduced via the Heretic library.

## 🚀 Overview

Abliterated version of [empero-ai/Qwythos-9B-Claude-Mythos-5-1M](https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M), a 9B reasoning & creative model on a Qwen3.5-9B base, post-trained on Claude Mythos & Fable traces, with a **1M-token context** (YaRN). The base is already largely uncensored; this build applies [Heretic](https://github.com/p-e-w/heretic) to trim residual refusals while preserving its style and coherence. Thinking model (`<think>`), function-calling capable.

## 📊 Abliteration Results

| Metric                                            | Value  |
| ------------------------------------------------- | ------ |
| Refusals after abliteration (harmful-prompt eval) | 53/100 |
| KL Divergence                                     | 0.0066 |

KL ≈ 0.007 → negligible capability change; targets the residual refusals left on an already-uncensored base.

## 🎯 Key Features

- **Claude-Mythos** creative & roleplay style
- Reasoning / thinking mode, native function calling
- **1M-token context** (YaRN)
- Further-reduced refusals via Heretic

## 🏷️ Available Versions

| Tag             | Size    | BPW  | Notes              |
| --------------- | ------- | ---- | ------------------ |
| IQ3_M           | 4.4 GB  | 3.66 | Smallest           |
| IQ4_XS          | 5.2 GB  | 4.25 | Great quality/size |
| latest / Q4_K_M | 5.6 GB  | 4.85 | **Recommended**    |
| Q5_K_M          | ~6.5 GB | 5.68 | Higher quality     |
| Q8_0            | ~9.5 GB | 8.5  | Near-lossless      |

## 💻 Quick Start

```bash
ollama run richardyoung/qwythos-9b-abliterated
```

## 🛠️ Use Cases

- Creative writing, roleplay, long-context fiction
- Reasoning, agentic / function-calling tasks

## 🔧 Technical Details

- **Base Model:** [empero-ai/Qwythos-9B-Claude-Mythos-5-1M](https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M) (Qwen3.5-9B, `qwen3_5`, 1M context)
- **Abliteration:** [Heretic](https://github.com/p-e-w/heretic), Trial 104 (53/100 refusals, KL 0.0066)
- **Quantization:** GGUF via llama.cpp (text generation)

## ⚠️ Disclaimer

This model has reduced safety guardrails and will engage with a wider range of prompts. Use responsibly and in accordance with applicable laws.

## 🙏 Acknowledgments

- **Base Model:** empero-ai (Qwythos / Claude Mythos series)
- **Abliteration:** [Heretic](https://github.com/p-e-w/heretic) by p-e-w
- **Quantization:** [llama.cpp](https://github.com/ggml-org/llama.cpp)

---

*Built & maintained by [Richard Young](https://deepneuro.ai/richard) · DeepNeuro*

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)