Language models with refusal behaviors removed.
Removes refusal directions from a language model, checks its capabilities, and saves the modified model.
When to use it
Use to remove a language model's refusal behaviors or analyze its refusal geometry.
Give it a model ID or local path; it modifies the model and returns capability metrics and the saved model's location.
What you provide
Requires Python 3.10 or later.
Requires PyTorch 2.1 or later; CUDA is recommended, but CPU operation is supported.
This skill
OBLITERATUS
Installs obliteratus
Sends anonymous research telemetry
HuggingFace
Loads models from Hugging Face