Safety Edits That Don't Stick: datasets ValueShift Research, Lexsi hackathon: RMU unlearning and Antidote repair on Tiny Aya. Repair artifacts are deliberately harmful. ValueShift/eoh-results Updated 4 days ago • 69 ValueShift/exp1-antidote-logs Updated 4 days ago • 181 ValueShift/exp1-antidote-ru-logs Updated 4 days ago • 94 ValueShift/exp1-antidote-adaptive-logs Updated 4 days ago • 100
Safety Edits That Don't Stick: models ValueShift Research, Lexsi hackathon: RMU unlearning and Antidote repair on Tiny Aya. Repair artifacts are deliberately harmful. ValueShift/exp1-antidote-adapters Updated 4 days ago ValueShift/exp1-antidote-ru-adapters Updated 4 days ago ValueShift/exp1-antidote-adaptive-adapters Updated 4 days ago
Safety Edits That Don't Stick: datasets ValueShift Research, Lexsi hackathon: RMU unlearning and Antidote repair on Tiny Aya. Repair artifacts are deliberately harmful. ValueShift/eoh-results Updated 4 days ago • 69 ValueShift/exp1-antidote-logs Updated 4 days ago • 181 ValueShift/exp1-antidote-ru-logs Updated 4 days ago • 94 ValueShift/exp1-antidote-adaptive-logs Updated 4 days ago • 100
Safety Edits That Don't Stick: models ValueShift Research, Lexsi hackathon: RMU unlearning and Antidote repair on Tiny Aya. Repair artifacts are deliberately harmful. ValueShift/exp1-antidote-adapters Updated 4 days ago ValueShift/exp1-antidote-ru-adapters Updated 4 days ago ValueShift/exp1-antidote-adaptive-adapters Updated 4 days ago