[InternVideo1-B] Evaluation Not Triggered During Training with Specified Interval

Dear InternVideo Group,

Thank you for your wonderful work!

I encountered an issue while attempting to finetune the InternVideo1-B model. According to the config instructions, I set the 'evaluation' as follows:

evaluation = dict( interval=5, metrics=['top_k_accuracy', 'mean_class_accuracy'], topk=(1, 5) )

This setting should trigger evaluation after every 5 epochs during training. However, based on my logs, it appears that the evaluations are not being conducted as expected. Specifically, the log output indicates that only the training process is taking place, with no records of evaluation activities.

Could you please advise if there's anything incorrect in my config? Additionally, any guidance on what might be causing this discrepancy and how to resolve it would be greatly appreciated.

Thank you in advance for your help.

Start by locating the finetuning entry point and the configuration containing evaluation = dict(interval=5, metrics=['top_k_accuracy', 'mean_class_accuracy'], topk=(1, 5)). Check how the training loop consumes this setting and compare the run logs with the expected evaluation schedule. Done means evaluation runs every five epochs and records the requested metrics.