The file size of best_metric_epoch_N.pth and that of epoch_N.pth are different

[myfrannie]

Branch

main branch (mmpretrain version)

Describe the bug

In order to save the best checkpoint, I set default hook config as checkpoint=dict(type='CheckpointHook', interval=10, save_best='accuracy/top1').
Then below is the result

-a---        2025-02-12  PM 3:20       12351503 best_accuracy_top1_epoch_118.pth
-a---        2025-02-12  AM 8:44       22270961 epoch_100.pth
-a---        2025-02-12 PM 12:06       22445297 epoch_110.pth
-a---        2025-02-12 PM 12:48       22490801 epoch_112.pth
-a---        2025-02-12  PM 3:20       22596081 epoch_118.pth
-a---        2025-02-12  PM 4:23       22620529 epoch_120.pth
-a---        2025-02-12  PM 7:54       22794865 epoch_130.pth
-a---        2025-02-12 PM 11:13       22968305 epoch_140.pth
-a---        2025-02-13  AM 2:33       23142449 epoch_150.pth

Why the file size of best_accuracy_top1_epoch_118.pth is different from that of epoch_118.pth ??
Is there any difference between saving the last checkpoint and saving the best checkpoint?
The inference results are the same in both cases.