reinthal/dev-instructed-deception-gemma-3-27b-it-None-relabel-v5
dev-instructed-deception-gemma-3-27b-it-None-relabel-v5 — v5 relabel + split Copy of aletheias-quest/dev-instructed-deception-gemma-3-27b-it-None with the v5 belief-relative label (see reinthal/aletheias-dev-relabel-v5 for the method: 20x neutral resample -> DeepSeek-V4-Flash judge, no canonicalization) and a train/test/validation split column. Added columns: deceptive (v5 label; official fallback where excluded), official (original dev label), relabeled (v5 != official)… See the full description on the dataset page: https://huggingface.co/datasets/reinthal/dev-instructed-deception-gemma-3-27b-it-None-relabel-v5.
dev-instructed-deception-gemma-3-27b-it-None-relabel-v5 — v5 relabel + split
Copy of aletheias-quest/dev-instructed-deception-gemma-3-27b-it-None with the v5 belief-relative label (see reinthal/aletheias-dev-relabel-v5 for the method: 20x neutral resample -> DeepSeek-V4-Flash judge, no canonicalization) and a train/test/validation split column.
Added columns: deceptive (v5 label; official fallback where excluded), official (original dev label), relabeled (v5 != official), excluded (belief unestablished), split (train/test/validation = 75/15/10, assigned by md5 of the question text so repeated questions never straddle splits — consistent across all 21 -relabel-v5 sets).