ResearchDatasetsQwen2.5-3B Layer 12 activationsQwen2.5-3B Layer 12 activationsFollowBenchmarksTaskDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyTaskDatasetSOTA resultTrendResultsSparse AutoencodingQwen2.5-3B Layer 12 activations (held-out)48.9Relative Error4