Why format reward is always zero?

[amoreZgx1n]

Thanks for you excellent work! I train RL for 20 steps using your cold-start model, but the fornat reward is always 0.