tril

nvtripy.tril(tensor: Tensor, diagonal: int = 0) Tensor[source]

Returns the lower triangular part of each \([M, N]\) matrix in the tensor, with all other elements set to 0. If the tensor has more than two dimensions, it is treated as a batch of matrices.

Parameters:
  • tensor (Tensor) – [dtype=T1] The nvtripy tensor to operate on.

  • diagonal (int) –

    The diagonal above which to zero elements. diagonal=0 indicates the main diagonal which is defined by the set of indices \({{(i, i)}}\) where \(i \in [0, min(M, N))\).

    Positive values indicate the diagonal which is that many diagonals above the main one, while negative values indicate one which is below.

Returns:

[dtype=T1] A tensor of the same shape as this tensor.

Return type:

Tensor

DATA TYPE CONSTRAINTS:
Example: Main Diagonal
1input = tp.iota((2, 1, 3, 3), dim=2) + 1.0
2output = tp.tril(input)
Local Variables
>>> input
tensor(
    [[[[1, 1, 1],
       [2, 2, 2],
       [3, 3, 3]]],


     [[[1, 1, 1],
       [2, 2, 2],
       [3, 3, 3]]]], 
    dtype=float32, loc=gpu:0, shape=(2, 1, 3, 3))

>>> output
tensor(
    [[[[1, 0, 0],
       [2, 2, 0],
       [3, 3, 3]]],


     [[[1, 0, 0],
       [2, 2, 0],
       [3, 3, 3]]]], 
    dtype=float32, loc=gpu:0, shape=(2, 1, 3, 3))
Example: Two Diagonals Above Main
1output = tp.tril(input, diagonal=2)
Local Variables
>>> input
tensor(
    [[1, 1, 1, 1, 1],
     [2, 2, 2, 2, 2],
     [3, 3, 3, 3, 3],
     [4, 4, 4, 4, 4],
     [5, 5, 5, 5, 5]], 
    dtype=float32, loc=gpu:0, shape=(5, 5))

>>> output
tensor(
    [[1, 1, 1, 0, 0],
     [2, 2, 2, 2, 0],
     [3, 3, 3, 3, 3],
     [4, 4, 4, 4, 4],
     [5, 5, 5, 5, 5]], 
    dtype=float32, loc=gpu:0, shape=(5, 5))
Example: One Diagonal Below Main
1output = tp.tril(input, diagonal=-1)
Local Variables
>>> input
tensor(
    [[1, 1, 1, 1, 1],
     [2, 2, 2, 2, 2],
     [3, 3, 3, 3, 3],
     [4, 4, 4, 4, 4],
     [5, 5, 5, 5, 5]], 
    dtype=float32, loc=gpu:0, shape=(5, 5))

>>> output
tensor(
    [[0, 0, 0, 0, 0],
     [2, 0, 0, 0, 0],
     [3, 3, 0, 0, 0],
     [4, 4, 4, 0, 0],
     [5, 5, 5, 5, 0]], 
    dtype=float32, loc=gpu:0, shape=(5, 5))