Downloads · 30 days
155
21% of all-time downloads
mythos-9b-merged-heretic
richardyoung/mythos-9b-merged-heretic
mythos-9b-merged-heretic is a text generation model from richardyoung. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as apache-2.0.
[!TIP] This model is reproducible! See the README in the reproduce directory for more information.
All-time downloads
734
Public
Parameters
8.2B
16.4 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors16.4 GB · 100%
At a glance
- Task
- Text Generation
- Library
- transformers
- License
- apache-2.0
- Model type
- qwen3
- Access
- Public
- Created
- Jun 26, 2026
- Updated
- Sep 26, 2026
- SHA
- 4edcbfd1
Try a prompt
Base models
Model card
From the Hugging Face model README
This is a decensored version of King3Djbl/mythos-9b-merged, made using Heretic v1.4.0
[!TIP] This model is reproducible!
See the README in the
reproducedirectory for more information.
Abliteration parameters
| Parameter | Value |
|---|---|
| direction_index | 18.72 |
| attn.o_proj.max_weight | 1.21 |
| attn.o_proj.max_weight_position | 27.88 |
| attn.o_proj.min_weight | 1.02 |
| attn.o_proj.min_weight_distance | 20.65 |
| mlp.down_proj.max_weight | 1.02 |
| mlp.down_proj.max_weight_position | 30.52 |
| mlp.down_proj.min_weight | 0.74 |
| mlp.down_proj.min_weight_distance | 19.34 |
Performance
| Metric | This model | Original model (King3Djbl/mythos-9b-merged) |
|---|---|---|
| KL divergence | 0.0211 | 0 (by definition) |
| Refusals | 7/100 | 79/100 |
Mythos-9B
Uncensored fine-tuned agent model — Built from Qwen3-9B with 47,824 agent traces for tool use, shell commands, and reasoning. Features native thinking mode for complex multi-step tasks.
Quick Start
Ollama (Easiest)
ollama pull fableforge/mythos-9b
ollama run fableforge/mythos-9b
llama.cpp
./llama-cli -m mythos-9b-Q4_K_M.gguf -ngl 99
HuggingFace Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("King3Djbl/mythos-9b-merged", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("King3Djbl/mythos-9b-merged")
Performance (Q4_K_M on M3 Mac)
| Category | Score | Notes |
|---|---|---|
| Censorship Resistance | 4.5/5 | 9/10 categories: full answers |
| Tool-Use | 4.8/5 | Shell, code, Docker, K8s, SQL |
| Reasoning | 4.5/5 | Logic, debugging, system design |
| Speed (thinking) | 10.7 tok/s | With thinking tokens |
| Speed (no-think) | ~15-20 tok/s | Estimated |
Model Family
| Model | Size | Censorship | Best For |
|---|---|---|---|
| ShellWhisperer-1.5B | 1.5B | 3.5/5 | Shell/terminal, edge devices |
| Mythos-9B | 9B | 4.5/5 | General agent, tool calls, reasoning |
| Mythos-9B-Enhanced | 9B | 4.8/5 | Agent + security research |
| Mythos-9B-Unhinged | 9B | 5/5 | Fully uncensored |
Training
Fine-tuned on FableForge Mix A dataset (47,824 examples) — agent traces, shell commands, code generation, and multi-step reasoning. 98.3% of the 2.8M formatted examples remain untapped for future training iterations.
License: Apache 2.0
Built & maintained by Richard Young · DeepNeuro
- Task
- Text Generation
- Library
- transformers
- Type
- qwen3
- License
- apache-2.0
- Languages
- en
- Created
- Jun 26, 2026
- Updated
- Sep 26, 2026