richardyoung/ fable-qwen2.5-3b-abliterated:latest

255 Downloads Updated 1 month ago

Abliterated Fable-Qwen2.5-3B: a small agentic Qwen2.5-3B fine-tune decensored with Heretic. 3/100 refusals, KL 0.050.

tools

ollama run richardyoung/fable-qwen2.5-3b-abliterated

curl http://localhost:11434/api/chat \
  -d '{
    "model": "richardyoung/fable-qwen2.5-3b-abliterated",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='richardyoung/fable-qwen2.5-3b-abliterated',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'richardyoung/fable-qwen2.5-3b-abliterated',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 1 month ago

1 month ago

2b3e6bcf92a0 · 1.9GB

model

archqwen2

·

parameters3.09B

·

quantizationQ4_K_M

1.9GB

template

{{- if or .System .Tools }}<|im_start|>system {{- if .System }} {{ .System }} {{- end }} {{- if .Too

1.2kB

license

Apache License 2.0 - https://www.apache.org/licenses/LICENSE-2.0

64B

params

{ "num_ctx": 8192, "stop": [ "<|im_start|>", "<|im_end|>", "<|endoft

100B

Readme

fable-qwen2.5-3b-abliterated

Abliterated Fable-Qwen2.5-3B: a small agentic Qwen2.5-3B fine-tune decensored with Heretic. ³⁄₁₀₀ refusals, KL 0.050.

Warning: Abliterated model with reduced safety guardrails. It will comply with requests a safety-tuned model would refuse. You are responsible for how you use its outputs.

Run

ollama run richardyoung/fable-qwen2.5-3b-abliterated

Tag	Quant	Size	Context
`latest`, `Q4_K_M`	Q4_K_M	1.9 GB	32K native (8K default; raise with `/set parameter num_ctx`)

About

Fable-Qwen2.5-3B-Agentic is an Unsloth fine-tune of Qwen2.5-3B-Instruct aimed at agentic tasks.

Architecture: Qwen2.5 · 3.1B
Original model: devxyasir/fable-qwen2.5-3b-agentic-merged
Full weights: richardyoung/fable-qwen2.5-3b-agentic-merged-heretic
MLX: richardyoung/fable-qwen2.5-3b-abliterated-mlx

Best for: Lightweight local agent and chat on low-memory machines.

Abliteration

Made with Heretic v1.4.0.

Metric	This model	Original
Refusals	³⁄₁₀₀	⁹⁶⁄₁₀₀
KL divergence	0.0503	0

License

apache-2.0, inherited from the original model.

Built & maintained by Richard Young · DeepNeuro

# fable-qwen2.5-3b-abliterated

Abliterated Fable-Qwen2.5-3B: a small agentic Qwen2.5-3B fine-tune decensored with Heretic. 3/100 refusals, KL 0.050.

> **Warning:** Abliterated model with reduced safety guardrails. It will comply with requests a safety-tuned model would refuse. You are responsible for how you use its outputs.

## Run

```bash
ollama run richardyoung/fable-qwen2.5-3b-abliterated
```

| Tag | Quant | Size | Context |
| :-- | :-- | :-- | :-- |
| `latest`, `Q4_K_M` | Q4_K_M | 1.9 GB | 32K native (8K default; raise with `/set parameter num_ctx`) |

## About

Fable-Qwen2.5-3B-Agentic is an Unsloth fine-tune of Qwen2.5-3B-Instruct aimed at agentic tasks.

**Architecture:** Qwen2.5 · 3.1B  
**Original model:** [devxyasir/fable-qwen2.5-3b-agentic-merged](https://huggingface.co/devxyasir/fable-qwen2.5-3b-agentic-merged)  
**Full weights:** [richardyoung/fable-qwen2.5-3b-agentic-merged-heretic](https://huggingface.co/richardyoung/fable-qwen2.5-3b-agentic-merged-heretic)  
**MLX:** [richardyoung/fable-qwen2.5-3b-abliterated-mlx](https://huggingface.co/richardyoung/fable-qwen2.5-3b-abliterated-mlx)

**Best for:** Lightweight local agent and chat on low-memory machines.

## Abliteration

Made with [Heretic](https://github.com/p-e-w/heretic) v1.4.0.

| Metric | This model | Original |
| :-- | :--: | :--: |
| Refusals | 3/100 | 96/100 |
| KL divergence | 0.0503 | 0 |

## License

apache-2.0, inherited from the original model.

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

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