OBLITERATUS open-source toolkit targets LLM refusal behavior
OBLITERATUS is an open-source toolkit for analyzing and removing refusal-related internal representations in large language models without retraining or fine-tuning. It provides extraction methods including PCA, mean-difference, sparse autoencoder decomposition and whitened SVD, along with a Gradio interface on Hugging Face Spaces and a Python API. The project claims that runs with telemetry enabled contribute anonymous benchmark data. Its claims of superiority and preservation of core language capabilities remain unverified in the supplied report.
The toolkit gives researchers an inspectable, reproducible workflow for studying and modifying refusal directions in LLMs, with multiple analysis methods, APIs and an interactive interface.
OBLITERATE THE CHAINS THAT BIND YOU

https://huggingface.co/spaces/pliny-the-prompter/
O B L I T E R A T U S
Break the chains. Free the mind. Keep the brain.
Some source content is omitted. Continue reading on the original site.