iree-turbine assert in _to_meta_tensor

[FlintWangacc]

I am debugging a module with following code snippet.

class StateUpdateModule(CompiledModule):
    ....
   def run_initialize(
         self, x=AbstractTensor(BATCH_SIZE, None, dtype=torch.int64)
    ):
   ....

It seems iree-turbine has some trouble deal with AbstractTensor(BATCH_SIZE, None, dtype=torch.int64).
It will assert when iree-turbine try to compile this code snippet.
It will output the following message.

....
File "iree/turbine/aot/support/procedural/base.py", line 135, in __call__
return current_ir_trace().handle_call(self, args, kwargs)
File "iree/turbine/aot/support/procedural/tracer.py" line 150, in handle_call
return target.resolve_call(self, *args, **kwargs)
File "iree/turbine/aot/builtins/jittable.py", line 187, in resolve_call
ir_arg, pytorch_arg = self._split_py_arg(py_arg)
File "iree/turbine/aot/builtins/jittable.py", line 311, in _split_py_arg
meta_tensor = arg._to_meta_tensor()
File "/iree/turbine/aot/support/procedural/primitives.py", line 209, in _to_meta_tensor
assert not any(
AssertionError: Unsupported dynamic dims in meta tensor

[FollowHeart007]

I encountered the same issue while trying to deploy the Llama model.