As Matthias answered above, there are likely 0s in your arrays that are turned into Nans by np.divide(). This is because the default replacement for Nan in np.nan_to_num() is 0 and the log2 of 0 returns Inf.

Try this instead:

l2fc = np.log2(np.nan_to_num(np.divide(a, b), nan=1))

The most likely explanation is, that your data contains zeros. This would return Inf, since you can't compute a log value of 0. The typical approach is, that pseudo counts are added prior to calculating the log, e.g. +1.