Request for Random Seeds: ResNet-20 on CIFAR-100 Reproduction (Table 3) #2
Description
Hello authors,
Thank you for open-sourcing this impressive work. I am currently trying to reproduce the results reported in Table 3 of the paper.
Specifically, I am focusing on the ResNet-20 on CIFAR-100 experiments:
·MCNC (w/o LoRA): Reported accuracy ~34.7%
·MCNC (w/ LoRA): Reported accuracy ~36.7%
In my reproduction attempts using the provided scripts, I noticed that the final accuracy is quite sensitive to the initialization of the random generator. My results tend to fluctuate around 33.5% ~ 34.3%, and it is difficult to consistently match the reported baseline performance.
Could you please share the specific random seeds used to achieve the results in Table 3? Having a fixed seed that guarantees the reported baseline performance would be extremely helpful for debugging and ensuring my experimental setup is correct.
Thank you for your time and help!