In lesson 2 at, he simply chances cycle_mult to 2. cycle_mult makes the learning rate now decrease more slowly by a factor of 2 so why does the number of number of test runs double?

Look at training_phase.ipynb in the dl2 folder. Inside there is a SGDR part. It explains how it works and also allows you to plot what is happening to the learning rate as you change cycle_len and cycle_mult.