Effect of random seed on experimental results

I was interested in your work, so I downloaded the code to try it out. However, under the Settings of M=16, Shot=5, CTP=middle, the test acc of the three random seeds is very different, respectively: 67.7%，73.3%，64.0%。What might be the reason for this? As a rule of thumb, random seeds should not have such a large effect on experimental performance. 
Looking forward to your reply！