about RL

[linhan12]

Hi, I want to ask how you apply GRPO training on VLAA-Thinking-RL data which contains reasoning process. Are the thinking parts considered into the reward calculation?

[hieuhthh]

I am wondering about this too!

[ImKeTT]

Thank you for your interests in our work and apologize for the late reply. When applying RL, we did not consider rewards on the thinking process, we still employed the final answer rewards but through a mixed verifier.