How to reproduce experimental results

[ReCodeLife]

I have trained on a server and obtained a checkpoint. However, when I tried to reproduce the results on my local machine, I found a significant difference in accuracy. So, I attempted to reproduce the effect with a minimal experiment.
When training on my local machine

bash tools/benchmarks/classification/mim_dist_train.sh myce/maedownstream/dataset_right_style-label_mae_1xb16.py /home/dirichlet/data/PycharmProjects/data/checkpoint/mae-75-right-t3-b100-pretrain/8-4-19-15/work_dirs/myVideoMae/exec_mae_1xb16-steplr-50e_style/epoch_100.pth

I achieved an accuracy/top1 of 88.4281.

#some output
09/19 14:13:28 - mmengine - INFO - Exp name: mae-right-p16_linear-8xb128-coslr-90e_style_20230919_140433
09/19 14:13:28 - mmengine - INFO - Saving checkpoint at 1 epochs
09/19 14:13:36 - mmengine - INFO - Epoch(val)   [1][100/633]    eta: 0:00:36  time: 0.0563  data_time: 0.0004  memory: 2750  
09/19 14:13:43 - mmengine - INFO - Epoch(val)   [1][200/633]    eta: 0:00:28  time: 0.0630  data_time: 0.0004  memory: 1701  
09/19 14:13:49 - mmengine - INFO - Epoch(val)   [1][300/633]    eta: 0:00:21  time: 0.0750  data_time: 0.0138  memory: 1701  
09/19 14:13:55 - mmengine - INFO - Epoch(val)   [1][400/633]    eta: 0:00:14  time: 0.0648  data_time: 0.0050  memory: 1701  
09/19 14:14:01 - mmengine - INFO - Epoch(val)   [1][500/633]    eta: 0:00:08  time: 0.0565  data_time: 0.0003  memory: 1701  
09/19 14:14:08 - mmengine - INFO - Epoch(val)   [1][600/633]    eta: 0:00:02  time: 0.0596  data_time: 0.0003  memory: 1701  
09/19 14:14:09 - mmengine - INFO - Epoch(val) [1][633/633]  accuracy/top1: 88.4281
09/19 14:14:11 - mmengine - INFO - The best checkpoint with 88.4281 accuracy/top1 at 1 epoch is saved to best_accuracy/top1_epoch_1.pth.

But when I use test code multiple times

bash tools/benchmarks/classification/mim_dist_test.sh /home/dirichlet/data/PycharmProjects/mmselfsup/myce/maedownstream/mae-right-p16_linear-8xb128-coslr-90e_style.py /home/dirichlet/data/PycharmProjects/mmselfsup/work_dirs/mae-right-p16_linear-8xb128-coslr-90e_style/best_accuracy/top1_epoch_1.pth

I have always obtained an accuracy/top1 of 88.4722.
Although the change is small, it does present different results. Is this reasonable, or did I make a mistake?