[Bug] Your pre -training model accuracy is different from logs

[veizei-K]

Branch

mmcls-1.x branch (v1.0.0rc6 or other 1.x version)

Describe the bug

https://github.com/open-mmlab/mmpretrain/blob/main/configs/mobilenet_v3/README.md
The accuracy of the mobilenet-v3-large_8xb128_in1k model on this page is 73.49 and 91.31. However, the training records in the logs do not reflect this. Additionally, there are discrepancies between the configuration file and the training logs. For instance, the configuration file specifies validation after every epoch, but the logs show validation occurring every ten epochs.

Environment

None

Other information

No response