Results vary greatly across experiments

[William-HYWu]

Hi. Thanks for your wonderful works! I'm trying mamba block on several datasets, but the results seem to vary greatly even with the same configuration (the accuracy can shift by 2% or more). I've already set the seed with the following code, did I miss anything?

random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)