The issue about the batch_size and training step

[FFppran]

My GPU memory only have 4G(gtx 1050ti). if I set batch_size=32,it will prompt Memory Error.So,do I need to increase the number of training steps to supplement the lack of training samples?

[LevinJ]

Hi @FFppran , I think you might try training with current iteration numbers configuration and see how it goes, it's possible that the chage of batch_size has no effect on the final result, as the training is already fully done (both train and test error pleateaued in currentlearning chart)