Setting up the training configuration

I am using an RTX 3090 for training, but due to hardware limitations, the maximum batch_size I can set is only 2. Under this setting, how should I adjust other configurations to achieve the results mentioned in your article?