Please fix this wrong

First of all:
my env:
MacBookAir M1
tensorflow 0.1a3

I want to study TensorFlow by using this code and datasets, so I copied this code from this link(and I didn’t change any code in it.):
https://www.kaggle.com/roblexnana/cifar10-with-cnn-for-beginer/notebook

Problem:
At first, the accuracy is 0.2 which is normal, and the accuracy increase as training which is normal too.
when it comes to around 10/100 epoches, the accuracy began to wane, and about at 20/100 epoches, the accuracy down to ~0.1.
and I put it into my winodws computer, using tensorflow 2.7, the code performs as well as the link above(accuracy: 0.80+).

but I cannot send an issue in Github, so I put this problem here hope somebody matters will see this.

What is tensorflow 0.1a3?

I am using MacBook Air (M1), the version of tensorflow on this M1 is 0.1 alpha 3.
the pip list outputs:

tensorboard             2.4.0
tensorboard-data-server 0.6.1
tensorboard-plugin-wit  1.8.0
tensorflow-addons       0.1a3
tensorflow-estimator    2.3.0
tensorflow              0.1a3

I encountered this problem with this version.
I submit some email on apple’s website, they said they will call me back within 3 days.

Can you follow the Apple silicon install guide: