Batch size doesn't affect

[momendoufu]

In my experiments, the evaluation accuracy get worse results when using big batch size. For example, at default setting, batch size 256 get 79.23% yet 512 get 78.38%, and it same as resnet18 and 50. I think it contray to paper's result of simclr.
Does anyone have same issue?

[leftthomas]

@momendoufu you could try different random seed and big dataset, such as ImageNet