List your company

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

0.738Sentence BLEU-1Trajectory-Aware Process Supervision
0.660.680250.70050.72075Apr 10, 2026

Evaluation results

MethodLinks
Trajectory-Aware Process SupervisionReward Configuration=Rform + Rexact + RDTW2026.040.738
GRPO (Rform + Rexact)Reward Configuration=Rform + Rexact2026.040.712
GRPO (Rform)Reward Configuration=Rform2026.040.666
SFTTraining Type=Supervised Fine-Tuning2026.040.663