Lexsi: Unlearning vs. Hiding
Collection
Public Lexsi artifacts: four RMU Tiny Aya checkpoints, eval traces, and the live unlearning-vs-hiding demo. • 6 items • Updated
Research checkpoints, evaluation data, and demos, including Lexsi and Apart.
Note Per-project collection: RMU checkpoints, eval data, and live demo
Note Research artifact: RMU-unlearned Tiny Aya (global)
Live · RMU audit: does unlearning hide knowledge?
Note Updated live version — research demo (precomputed traces)
Note Evaluation dataset, traces, and splits
Note Research artifact: RMU-unlearned Tiny Aya (earth)
Note Research artifact: RMU-unlearned Tiny Aya (fire)
Note Research artifact: RMU-unlearned Tiny Aya (water)
Note Research dataset: Phi-3.5 abstention / unanswerable-question probes
Note Per-project collection: Lexsi evaluation datasets