dataset
active
dataset:wikitext
Favorite
WikiText
Unlearning retain set.
Neighborhood — ranked by edge-count
Papers (1)
paper
SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability
mentions
Methods (1)
method
Unlearning evaluation
uses
Conditional negative clamping on WMDP-bio with MMLU retention.