Forward reward always 0

[ZhaoyangLi-1]

I run grpo_classification.py with my own dataset, while my code is always 0. How ot fix.

I use Qwen3-VL-8B-Instruct as the base model