torch.logsumexp — PyTorch 2.7 documentation (original) (raw)

torch.logsumexp(input, dim, keepdim=False, *, out=None)¶

Returns the log of summed exponentials of each row of the inputtensor in the given dimension dim. The computation is numerically stabilized.

For summation index jj given by dim and other indices ii, the result is

logsumexp(x)i=log⁡∑jexp⁡(xij)\text{logsumexp}(x)_{i} = \log \sum_j \exp(x_{ij})

If keepdim is True, the output tensor is of the same size as input except in the dimension(s) dim where it is of size 1. Otherwise, dim is squeezed (see torch.squeeze()), resulting in the output tensor having 1 (or len(dim)) fewer dimension(s).

Parameters

• input (Tensor) – the input tensor.
• dim (int or tuple of ints , optional) – the dimension or dimensions to reduce. If None, all dimensions are reduced.
• keepdim (bool) – whether the output tensor has dim retained or not.

Keyword Arguments

out (Tensor, optional) – the output tensor.

Example:

a = torch.randn(3, 3) torch.logsumexp(a, 1) tensor([1.4907, 1.0593, 1.5696]) torch.dist(torch.logsumexp(a, 1), torch.log(torch.sum(torch.exp(a), 1))) tensor(1.6859e-07)