Summation of a tensor is giving -inf as output even though no element has -inf value in it

data
1

The size of tensor is [4,884736].

I am doing the summation with torch.sum(dim=1)

I have checked whether the tensor has float(‘-inf’) value with :-
1] torch.where(tensor == float(‘-inf’), 1, 0) → count the non_zero → output is zero
2] torch.where(tensor < -7, 1, 0) → count the non_zero → output is zero. Thus specifying the no element is lesser than -7
3] tensor[tensor == float(‘-inf’)] == 0. Still the result is -inf.

Kindly suggest.

2

Could you post the min() and max() values of this tensor you are trying to reduce?

3

Here are the max() and min() value of the tensor obtained with torch.max() and torch.min()
MAX_VALUE = -0.0002
MIN_VALUE = -4.9502

4

Thank you for replying.
This issue can be closed.
It was solved by converting the tensor, before summation, from torch.float16 to torch.float64.