abliteratedmodels.org

Abliteration (directional ablation) models

Abliteration (directional ablation) is one of several refusal-removal techniques applied to open-weight models. The single residual-stream direction that mediates refusal is identified from harmful/harmless prompt pairs and projected out of the model weights. 191 models indexed here were produced this way.

191 models13.5M downloads

What this technique does

The single residual-stream direction that mediates refusal is identified from harmful/harmless prompt pairs and projected out of the model weights.

Showing the 120 most-downloaded of 191 models in this group. The remaining 71 are reachable through the full paginated index and are all listed in the sitemap.