Schismatic
AI & ML interests
Uncensorship, abiliteration
Recent Activity
Schismatic
A free software AI model uncensorship tool.
What is Schismatic?
Schismatic is a free software toolkit that takes an existing open-weights language model and surgically removes its built-in βI canβt help with thatβ refusals β without retraining the model, changing its facts, or degrading its writing quality.
This Hugging Face organization is the official home of the modified models produced by the Schismatic tool. If you donβt want to run the tool yourself, you can simply download a published model from here and use it directly.
What it does (and doesnβt)
- β Finds the small internal direction inside a transformer model that most strongly predicts a refusal.
- β Subtracts that direction from a handful of layers β a tiny, reversible, surgical edit.
- β Measures the result against the original under a fixed evaluation configuration (refusal rate, KL divergence, next-token entropy, coherence gate).
- β It does not retrain the model.
- β It does not change the modelβs knowledge or writing quality β independent measurements show the modified model stays very close to the original on normal tasks.
How it works, briefly
- Probe β feed the model a batch of harmless and harmful prompts and watch which inner signals fire on the harmful ones.
- Find the direction β identify the single direction inside the model that most strongly predicts a refusal.
- Subtract it β apply a small, surgical edit that removes that direction from a handful of internal layers.
- Measure β score the modified model against the original to confirm it still works correctly on normal tasks.
Supported model families
Recipes ship for:
- Qwen, Qwen3, Qwen3-MoE, Qwen3.6
- Gemma 3, Gemma 4
- GLM-5
- DeepSeek V4
See the benchmarks page for the latest numbers and the studies page for the published hyperparameter searches.
Who itβs for
Researchers, developers, and curious people who want to:
- Study how refusal behavior is encoded in modern LLMs.
- Use a model that simply answers the question β for security research, medicine, history, fiction, or any sensitive topic legitimately blocked by over-broad safety training.
- Reproduce, extend, or audit the modifications openly.
Install the tool
pip install schismatic
Then point it at any supported open-weights checkpoint to produce your own uncensored variant on your own hardware.
Project links
| π Website | schismatic.ai |
| π¦ PyPI package | pypi.org/project/schismatic |
| π» Source code | spacecruft.org/schismatic/schismatic |
| π€ Released models | huggingface.co/schismatic/models |
| π Benchmarks | schismatic.ai/benchmarks |
| π¬ Studies | schismatic.ai/studies |
| π Changelog | schismatic.ai/changelog |
| π Credits | schismatic.ai/credits |
License
Schismatic β including all code, recipes, study artifacts, and the modified models published under this organization β is licensed under the GNU Affero General Public License, version 3 or later (AGPL-3.0-or-later).