List your company

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

75.8Sentence BLEU-1Trajectory-Aware Process Supervision
70.28871.71973.1574.581Apr 10, 2026

Evaluation results

MethodLinks
Trajectory-Aware Process SupervisionReward Configuration=Rform + Rexact + RDTW2026.0475.8
GRPO (Rform + Rexact)Reward Configuration=Rform + Rexact2026.0474
GRPO (Rform)Reward Configuration=Rform2026.0471.3
SFTTraining Type=Supervised Fine-Tuning2026.0470.5