How to evaluate the trained model?

[happened]

How can I evaluate the model after the training is complete? Could you please provide some information on the corresponding metrics and evaluation methods?

[36m(TaskRunner pid=32107)[0m ("Initial validation metrics: {'val/reward': 0.5210420841683366, " [36m(TaskRunner pid=32107)[0m "'val/mean_response_length': 33.122244488977955, 'val/sum_response_length': " [36m(TaskRunner pid=32107)[0m "57.701402805611224, 'val/turn_count': 1.5991983967935872}")

[ultmaster]

That really depends on your definition of agents.

For generally information on how to evaluate a trained model with VERL, please refer to https://github.com/volcengine/verl/issues/298