Regarding usage of parameters "num_iters_per_epoch" and "num_splits_asr" in ASR Task. about espnet

[sw005320]

To clarify, here are the example statistics: Total number of batches in the training data: 100,000 num_splits_asr: 10 num_iters_per_epoch: 10,000 With these settings, by the end of the 10th epoch, will the model have seen the entire training dataset? May be my question is too trivial. Can anyone give answer for the above question.

Yes, the model would see the entire training dataset, then.

[venkat01010]

Thank you, @sw005320, for your reply; it helps a lot. I have a follow-up question. Suppose a model has been trained for 5 epochs and is then interrupted. When resuming the training, will the model only see the remaining training data (which was not seen during the previous 5 epochs) for the next consecutive 5 epochs?