21 days ago
-
v3s: add FR validation sets (template control 300, held-out wording 500) + README
-
v3s README: list code folder
-
v3s: add LoRA SFT trainer code
-
v3s README: list env files
-
v3s: add conda env files for the LoRA SFT
-
v3s README: add LoRA SFT training plan (InternVL3.5-4B, Ovis2-4B)
-
Add PRIME v3s SFT training set (clean images, 500 matched pairs, short answers)
22 days ago
-
Add PRIME v3 SFT training set (1,000 rows: false-premise + matched true-premise twins)
28 days ago
-
Add eval1k_images.zip (AIpsych COCO, VG, POPE images)
-
Add eval1k: AIpsych COCO/VG 1k sets, probe, false-refusal and hallucination eval files, scorer, handoff
-
initial commit