Same code and experiment setup (except batch size and hardware) but can't achieve the performance as paper

[LOONAZeo]

Hi~ I download the code from github, but I can't achieve the results as your work.

My setup:
Dataset: Full dataset
Epoch = 15
Batch size: 12
Learning Rate: 1e-4, with per 3-epochs decay scheduler with a factor of 0.5
Optimizer: Adam
GPU: 1 RTX 5090

This is the results by using the same code as github with the experiment setup above.

Many thanks!