Return the (n, n) boolean mask where entry [i, j] is True exactly when position i is allowed to attend to position j, which for causal attention means j <= i.

For n = 3 that is a lower triangle including the diagonal. This is what stops a language model reading its own answer.

Input

3

Output

tensor([[ True, False, False],
        [ True,  True, False],
        [ True,  True,  True]])