Slice bug on LibTorch backend

[wcshds]

Here is the code.

type Backend = LibTorch;

let tensor = Tensor::<Backend, 1, Int>::from_data([1, 2, 3, 4, 5]);
let tensor = tensor
    .clone()
    .slice([0..3])
    .equal(tensor.clone().slice([2..5]));
println!("{}", tensor);

Sometimes it panics.

[wcshds]

It seems that calling the clone() again after the second slice() could serve as a temporary workaround for this issue.

type Backend = LibTorch;

let tensor = Tensor::<Backend, 1, Int>::from_data([1, 2, 3, 4, 5]);
let tensor = tensor
    .clone()
    .slice([0..3])
    .equal(tensor.clone().slice([2..5]).clone());
println!("{}", tensor);

[nathanielsimard]

I'm gonna work on this, it's probably at the source of some data corruption we have #877

[nathanielsimard]

@wcshds added the fix, can you validate if it works with your model?

[wcshds]

@nathanielsimard Thank you! Now it works.