symbolic cache policy can't handle string inputs properly.

[jjsjann123]

🐛 Bug

import thunder
import torch
 
device = "cuda"
 
def foo(b):
    # TODO: 'device=device' doesn't work for "symbolic values" cache policy
    a = torch.arange(24, device=device).reshape(3, 8)
    return a + b
 
jfoo = thunder.jit(foo, cache="symbolic values")
 
out = jfoo(5)

The script hits

  File "/volume/thunder_jit.py", line 9, in foo
    a = torch.arange(24, device=device).reshape(3, 8)
^^^^^^^^^^^^^^^^^^
  File "/opt/pytorch/lightning-thunder/thunder/core/interpreter.py", line 1302, in wrapping_wrapper
    res = ufn(*uargs, **ukwargs)
          ^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/pytorch/lightning-thunder/thunder/core/jit_ext.py", line 374, in wrapper
    return fn(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^
  File "/opt/pytorch/lightning-thunder/thunder/core/symbol.py", line 323, in __call__
    result = self.meta(*args, **kwargs)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/pytorch/lightning-thunder/thunder/core/langctxs.py", line 136, in _fn
    result = fn(*args, **kwargs)
             ^^^^^^^^^^^^^^^^^^^
  File "/opt/pytorch/lightning-thunder/thunder/torch/__init__.py", line 573, in arange
    device = maybe_get_default_device(device)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/pytorch/lightning-thunder/thunder/torch/__init__.py", line 130, in maybe_get_default_device
    return device or get_default_device()
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/pytorch/lightning-thunder/thunder/core/proxies.py", line 333, in __bool__
    raise NotImplementedError(f"__bool__ is not implemented for {type(self)}")
NotImplementedError: __bool__ is not implemented for <class 'thunder.core.proxies.StringProxy'>

inline device="cuda", the issue goes away. It also only errors out with symbolic values cache, which is not surprising since the other cache would hard code the string as a constant.