Better Supervised Fine-tuning for VQA: Integer-Only Loss

Re-assign community