List your company

Medical Visual Question Answering on Path-VQA (Sentence BLEU-1)

71.3Sentence BLEU-1Trajectory-Aware Process Supervision
50.70856.05461.466.746Apr 10, 2026

Evaluation results

MethodLinks
Trajectory-Aware Process SupervisionReward Configuration=Rform + Rexact + RDTW2026.0471.3
GRPO (Rform + Rexact)Reward Configuration=Rform + Rexact2026.0466.5
GRPO (Rform)Reward Configuration=Rform2026.0464.4
SFTTraining Type=Supervised Fine-Tuning2026.0451.5