How can I get the metrics of the evaluation step?

[zwverse]

The paper says there are two metrics evaluated:

1. the correctness of the natural language answer, evaluated by GPT-as-judge answer accuracy score
2. the accuracy of the grounding coordinates, measured by grounding IoU

Can you show how / where to retrieve these two metrics?

[UeFan]

https://github.com/eric-ai-lab/GRIT/blob/44699c9b03f5c1681f7b1d21c82fef8a1b023f0f/grpo-gr/GRPO_GRTrainer.py#L1144C16-L1145C17

This line shows where the results (of the metrics/rewards) are stored.