log_softmax¶

Computes \(\log(\text{softmax}(x))\) along a single dim.

ATen source: aten._log_softmax

Inputs¶

Name	Description
`input`	Tensor of any rank

Name	Type	Description
`axis`	`int`	Dimension along which `log_softmax` is computed
`version`	`int`	Composite op version

Name	Shape	Description
`output`	same as `input`	Same shape and dtype as `input`

fp16, fp32, bf16.

import torch
from torch.nn.functional import log_softmax

input = torch.randn(2, 3)
output = log_softmax(input, dim=1)