List your company

Visual Question Answering on ScienceQA (ASR, Coh, Rat, Acc, ASR-C)

4.1CoherenceReShift
2.58162.97583.373.7642Jul 1, 2026

Evaluation results

MethodLinks
ReShiftBase model=Qwen2.5-VL-7B2026.074.11004.17894
ReShiftBase model=InternVL3.5-8B2026.074.1994.31903
BaseBase model=Qwen2.5-VL-7B2026.073.95—4.4389—
BaseBase model=InternVL3.5-8B2026.073.93—4.3992—
RewriteBase model=InternVL3.5-8B2026.073.39903.477913
RewriteBase model=Qwen2.5-VL-7B2026.073.37773.357913
BadVisionBase model=InternVL3.5-8B2026.073.23393.34697
BadTokenBase model=InternVL3.5-8B2026.073.07922.588413
BadVisionBase model=Qwen2.5-VL-7B2026.072.72453.598117
BadTokenBase model=Qwen2.5-VL-7B2026.072.64843.288311