During inference, is the length of the input video sequence 100 or 60?

[lexilii]

The article mentions that the length of the training samples is 60. However, in the provided inference script, the length used is 100. So I'm confused.