Triu operator which can be executed via nvFuser

[protonu]

🚀 Feature

We want to implement a Triu operator which can be executed by nvFuser

Motivation

We want to improve performance via nvFuser execution

Additional context

We can implement Triu like the current implementation of Tril

lightning-thunder/thunder/torch/__init__.py

Line 2392 in c5c0b77

def tril(a: TensorLike, /, diagonal: int = 0, *, fill_value: None | Number = None) -> TensorLike:

Or we can execute if calling the Triu exposed in nvFuser
(NVIDIA/Fuser#3637)

cc @t-vi

[mruberry]

@beverlylytle — would you be interested in adding triu? It and tril are interesting operations

[protonu]

I added to PR (#1631) to address this issue.

[protonu]

Closed by PR #1631.
Thanks @mruberry @tfogal @beverlylytle @nikitaved !