training and evaluating details

[2561681244]

Hello! I noticed that the training and test datasets are identical when running bash examples/ttrl/Qwen2.5-Math/aime.sh. To ensure proper evaluation, the training data should be distinct from the test data. So I cannot understand.