p-e-w/heretic
Language model censorship removal tool
Heretic is a utility that automatically removes safety alignment from transformer-based language models using directional ablation and parameter optimization. It enables users to modify model behavior without requiring expensive post-training processes.
- Stars
- 32,300
- Stars gained in 7 days
- +600
- Contributors
- 35
- Forks
- 3,600
- Days trending in the last 30
- 7 days
- Mark
- Watch — 923 stars per contributor. An automatic mark on public metrics: above 500 is watch, above 2,000 possible hype (not a judgement of quality).