Group and find all values that belong to n unique maximum values

You can find unique values for each group, get the list with three max values and select rows which are in this list:

df.groupby('Input')['Font'].\
apply(lambda x: x[x.isin(np.sort(x.unique())[-3:])]).\
reset_index(level=0)

Output:

     Input  Font
6   132426    50
7   132426    50
8   132426    50
9   132426    38
10  132426    38
11  132426    30
12  132426    30
0   133217    30
1   133217    25
2   133217    25
3   133217    21