Different test accuracy with changed batch size in test set

Hi everyone!

I’ve just found out that when I increased the batch size value to calculate accuracy in my test set, the accuracy is changed. For example:

BatchSize Test Set 1 Test Set 2 Test Set 3 Test Set 4 Test Set 5 Test Set 6 Test Set 7
512 99.60 99.428 95.628 96.8666 95.749 90.133 93.34
1024 99.63 99.428 95.628 96.8499 95.716 90.116 93.38

Even model.eval() is all set, the accuracy has a small gap between each batch size value in some set.

Do you have any explanations for this?

Thanks,