Jordine/qwen2.5-32b-introspection-v4-suggestive_yesno
Updated
introspective-models-full-run
Full v4 ablation: 19 Qwen2.5-32B variants for steering-vector introspection finetuning.
-
-
Jordine/qwen2.5-32b-introspection-v4-neutral_moonsun
Updated -
Jordine/qwen2.5-32b-introspection-v4-neutral_redblue
Updated -
Jordine/qwen2.5-32b-introspection-v4-neutral_crowwhale
Updated -
Jordine/qwen2.5-32b-introspection-v4-vague_v1
Updated -
Jordine/qwen2.5-32b-introspection-v4-vague_v2
Updated -
Jordine/qwen2.5-32b-introspection-v4-vague_v3
Updated -
Jordine/qwen2.5-32b-introspection-v4-food_control
Updated -
Jordine/qwen2.5-32b-introspection-v4-no_steer
Updated -
Jordine/qwen2.5-32b-introspection-v4-deny_steering
Updated -
Jordine/qwen2.5-32b-introspection-v4-corrupt_25
Updated -
Jordine/qwen2.5-32b-introspection-v4-corrupt_50
Updated -
Jordine/qwen2.5-32b-introspection-v4-corrupt_75
Updated -
Jordine/qwen2.5-32b-introspection-v4-flipped_labels
Updated -
Jordine/qwen2.5-32b-introspection-v4-rank1_suggestive
Updated -
Jordine/qwen2.5-32b-introspection-v4-concept_10way_digit_r1
Updated -
Jordine/qwen2.5-32b-introspection-v4-concept_10way_digit_r16
Updated -
Jordine/qwen2.5-32b-introspection-v4-sentence_localization
Updated -
Jordine/qwen2.5-32b-introspection-v4-binder_selfpred
Updated