Zero sized tensor support for repeat_interleave

[zasdfgbnm]

Fixes #22753

[zasdfgbnm]

Test failure is unrelated

[zasdfgbnm]

@pytorchbot rebase this please

[mrshenli]

Thanks for the fix!

[facebook-github-bot]

@mrshenli has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[mrshenli]

@pytorchbot rebase this please

[facebook-github-bot]

@mrshenli has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[facebook-github-bot]

@mrshenli merged this pull request in 520982d.

[gchanan]

@zasdfgbnm is there a reason that this function (repeat(Tensor)) is exposed at all? It's not documented and it looks like an implementation detail.

[zasdfgbnm]

@gchanan It is actually documented

It is part of an implementation detail, but I feel it useful and it would be good to expose this as a shorter and faster way of doing torch.arange(n).repeat_interleave(repeats) when I originally implement torch.repeat_interleave.

My personal use case for this is:
I have a packed sequence s = tensor([x1, x2, ..., xn, y1, y2, ..., ym, ....]) and a tensor storing the size of each sequence counts = tensor([n, m, ...]), and I want to have a tensor of the sequence id for each element in s. In this case, I would do a torch.repeat_interleave(counts).