small fix: Index validator enable int64

[gs-olive]

Description

• Currently, pending the implementation of Refactor truncate_long_and_double in Dynamo #2590, constant index Tensors are type int64 before being frozen. This workaround enables valid usecases while Refactor truncate_long_and_double in Dynamo #2590 is being implemented

Type of change

• Bug fix (non-breaking change which fixes an issue)

Checklist:

• [ x ] My code follows the style guidelines of this project (You can use the linters)
• [ x ] I have performed a self-review of my own code
• [ x ] I have commented my code, particularly in hard-to-understand areas and hacks
• [ x ] I have made corresponding changes to the documentation
• [ x ] I have added tests to verify my fix or my feature
• [ x ] New and existing unit tests pass locally with my changes
• [ x ] I have added the relevant labels to my PR in so that relevant reviewers are notified

[peri044]

any reason why you explicitly set dynamic=False instead of None here ?

[peri044]

^ same question here

[gs-olive]

@peri044 - when dynamic=None, as is currently the (Default) case, torch.compile will insert SymInts upon detection of a new batch dimension, which we cannot handle currently, causing errors in the compiler. We currently specify it manually in the torch_tensorrt.compile convenience frontend, as here:

TensorRT/py/torch_tensorrt/_compile.py

Lines 251 to 254 in e38a7f3

	# TODO: Remove dynamic=False when SymInt Dynamic shape support is ready
	boxed_fn = torch.compile(
	module, backend=torch_tensorrt_backend, dynamic=False, options={**kwargs}
	)

When users call torch.compile directly, however, they need to specify dynamic=False to avoid errors.