richardyoung/ qwable-9b-abliterated:Q4_K_M

471 4 days ago

Abliterated build of Qwable-9B-Claude-Fable, a Claude-style creative and reasoning model (Qwen3.5-9B base) with refusals reduced via the Heretic library. 

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

Details

4 days ago

97b419451001 · 5.6GB ·

model
qwen35
·
8.95B
·
Q4_K_M
params
{ "num_ctx": 8192, "stop": [ "<|im_end|>" ] }

Readme

Qwable-9B-Claude-Fable-Abliterated

Abliterated build of Qwable-9B-Claude-Fable, a Claude-style creative and reasoning model (Qwen3.5-9B base) with refusals reduced via the Heretic library.

🚀 Overview

Abliterated version of empero-ai/Qwable-9B-Claude-Fable-5, a 9B creative and reasoning model on a Qwen3.5-9B base, post-trained on Claude Fable traces. Refusals were reduced with Heretic while preserving its style and coherence. It is a thinking model (emits <think> reasoning) and supports native function calling.

📊 Abliteration Results

Metric Value
Refusals after abliteration (harmful-prompt eval) 50⁄100
KL Divergence 0.0073

A KL divergence near 0.007 means negligible capability change.

🎯 Key Features

  • Claude-Fable creative and roleplay style
  • Reasoning and thinking mode, with native function calling
  • 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 and 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/qwable-9b-abliterated

🛠️ Use Cases

  • Creative writing, roleplay, and long-context fiction
  • Reasoning and agentic or function-calling tasks

🔧 Technical Details

⚠️ 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 (Qwable and the Claude Fable series)
  • Abliteration: Heretic by p-e-w
  • Quantization: llama.cpp

Built and maintained by Richard Young, DeepNeuro