richardyoung/ mythos-nano-abliterated

113 Downloads Updated 1 month ago

Abliterated Mythos-nano: a 3B math/code reasoning model (VibeThinker-3B fine-tune) decensored with Heretic. 13/100 refusals, KL 0.061.

tools

ollama run richardyoung/mythos-nano-abliterated

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

from ollama import chat

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

import ollama from 'ollama'

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

Applications

Claude Code

Claude Code ollama launch claude --model richardyoung/mythos-nano-abliterated

OpenCode

OpenCode ollama launch opencode --model richardyoung/mythos-nano-abliterated

Hermes Agent

Hermes Agent ollama launch hermes --model richardyoung/mythos-nano-abliterated

OpenClaw

OpenClaw ollama launch openclaw --model richardyoung/mythos-nano-abliterated

Models

Name

2 models

Size / Usage

Context

Input

mythos-nano-abliterated:latest

1.9GB · 128K context window · Text · 1 month ago

mythos-nano-abliterated:latest

1.9GB

128K

Text

mythos-nano-abliterated:Q4_K_M

1.9GB · 128K context window · Text · 1 month ago

mythos-nano-abliterated:Q4_K_M

1.9GB

128K

Text

Readme

mythos-nano-abliterated

Abliterated Mythos-nano: a 3B math/code reasoning model (VibeThinker-3B fine-tune) decensored with Heretic. ¹³⁄₁₀₀ refusals, KL 0.061.

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/mythos-nano-abliterated

Tag	Quant	Size	Context
`latest`, `Q4_K_M`	Q4_K_M	1.9 GB	128K native (8K default; raise with `/set parameter num_ctx`)

About

Mythos-nano is a 3B reasoning model fine-tuned from WeiboAI/VibeThinker-3B, strong on competition math and LeetCode-style coding. It was not trained for tool calling or agents.

Architecture: Qwen2 · 3.1B
Original model: squ11z1/Mythos-nano
Full weights: richardyoung/Mythos-nano-heretic
MLX: richardyoung/mythos-nano-abliterated-mlx

Best for: Math and competitive-programming reasoning. Not recommended for function calling or autonomous coding agents.

Abliteration

Made with Heretic v1.4.0.

Metric	This model	Original
Refusals	¹³⁄₁₀₀	⁷⁹⁄₁₀₀
KL divergence	0.0610	0

License

mit, inherited from the original model.

Built & maintained by Richard Young · DeepNeuro

# mythos-nano-abliterated

Abliterated Mythos-nano: a 3B math/code reasoning model (VibeThinker-3B fine-tune) decensored with Heretic. 13/100 refusals, KL 0.061.

> **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/mythos-nano-abliterated
```

| Tag | Quant | Size | Context |
| :-- | :-- | :-- | :-- |
| `latest`, `Q4_K_M` | Q4_K_M | 1.9 GB | 128K native (8K default; raise with `/set parameter num_ctx`) |

## About

Mythos-nano is a 3B reasoning model fine-tuned from WeiboAI/VibeThinker-3B, strong on competition math and LeetCode-style coding. It was not trained for tool calling or agents.

**Architecture:** Qwen2 · 3.1B  
**Original model:** [squ11z1/Mythos-nano](https://huggingface.co/squ11z1/Mythos-nano)  
**Full weights:** [richardyoung/Mythos-nano-heretic](https://huggingface.co/richardyoung/Mythos-nano-heretic)  
**MLX:** [richardyoung/mythos-nano-abliterated-mlx](https://huggingface.co/richardyoung/mythos-nano-abliterated-mlx)

**Best for:** Math and competitive-programming reasoning. Not recommended for function calling or autonomous coding agents.

## Abliteration

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

| Metric | This model | Original |
| :-- | :--: | :--: |
| Refusals | 13/100 | 79/100 |
| KL divergence | 0.0610 | 0 |

## License

mit, 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)