GitHub · HTML
AIAnytime/ablate
🔓 Ablate — directional ablation (abliteration) toolkit for open-source LLMs. Automatic censorship/refusal removal via residual-stream direction ablation, with KL-guided search, an LLM-judge harness, and one-call push to the Hub.
- Stars
- 8
- Forks
- 1
- Stars gained in 24 hours
- −14
- Forks gained in 24 hours
- −4
Topics: abliteration, activation-engineering, ai-safety, huggingface, interpretability, llm, mechanistic-interpretability, optuna, python, pytorch
Latest data point: Nov 21, 2026 · 8 stars · 1 forks