LayerFault/dataset-clean-balanced-control
Updated • 23
Synthetic poisoning, duplication, trigger, contamination and negative-control datasets.
Note LF-CH-DATA-0001: Dataset clean balanced control.
Note LF-CH-DATA-0011: Dataset eval answer canary overlap.
Note LF-CH-DATA-0002: Dataset exact duplicate concentration.
Note LF-CH-DATA-0007: Dataset fake credential pattern.
Note LF-CH-DATA-0012: Dataset high dup benign control.
Note LF-CH-DATA-0009: Dataset label flip minority.
Note LF-CH-DATA-0003: Dataset near duplicate concentration.
Note LF-CH-DATA-0005: Dataset rare trigger label correlation.
Note LF-CH-DATA-0010: Dataset trigger split columns.
Note LF-CH-DATA-0008: Dataset unsafe code text.
Note LF-CH-DATA-0004: Dataset url concentration.
Note LF-CH-DATA-0006: Dataset zero width trigger.