Hi, I wonder if there is a function that I can use to compute the min/max on a tensor of scalars.
Also I saw that in the documentation, the reduce() function has been used multiple times to compute the sum of a tensor. It seems that the function is iterating over all the scalars and use atomic_add to get the sum. Wouldn’t this kind of operation requires O(n) operations and might not be optimal for a GPU backend? Is this a deliberate design choice and is there a reason for this?