Reloading torch triggers RuntimeError

Question

When trying to reload the torch library in Python, by running the following lines of code, I encountered the error below:

import torch
import importlib
importlib.reload(torch)

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<ipython-input-5-6781a72fd549> in <cell line: 3>()
      1 import torch
      2 import importlib
----> 3 importlib.reload(torch)

6 frames
/usr/lib/python3.10/importlib/__init__.py in reload(module)
    167         if spec is None:
    168             raise ModuleNotFoundError(f"spec not found for the module {name!r}", name=name)
--> 169         _bootstrap._exec(spec, module)
    170         # The module may have replaced itself in sys.modules!
    171         return sys.modules[name]

/usr/lib/python3.10/importlib/_bootstrap.py in _exec(spec, module)

/usr/lib/python3.10/importlib/_bootstrap_external.py in exec_module(self, module)

/usr/lib/python3.10/importlib/_bootstrap.py in _call_with_frames_removed(f, *args, **kwds)

/usr/local/lib/python3.10/dist-packages/torch/__init__.py in <module>
   1951     from torch import compiler as compiler
   1952 
-> 1953     class _TritonLibrary:
   1954         lib = torch.library.Library("triton", "DEF")
   1955         ops_table: Dict[Tuple[str, str], Callable] = {}

/usr/local/lib/python3.10/dist-packages/torch/__init__.py in _TritonLibrary()
   1952 
   1953     class _TritonLibrary:
-> 1954         lib = torch.library.Library("triton", "DEF")
   1955         ops_table: Dict[Tuple[str, str], Callable] = {}
   1956 

/usr/local/lib/python3.10/dist-packages/torch/library.py in __init__(self, ns, kind, dispatch_key)
     62         frame = traceback.extract_stack(limit=3)[0]
     63         filename, lineno = frame.filename, frame.lineno
---> 64         self.m: Optional[Any] = torch._C._dispatch_library(kind, ns, dispatch_key, filename, lineno)
     65         self.ns = ns
     66         self._op_defs: Set[str] = set()

RuntimeError: Only a single TORCH_LIBRARY can be used to register the namespace triton; please put all of your definitions in a single TORCH_LIBRARY block.  If you were trying to specify implementations, consider using TORCH_LIBRARY_IMPL (which can be duplicated).  If you really intended to define operators for a single namespace in a distributed way, you can use TORCH_LIBRARY_FRAGMENT to explicitly indicate this.  Previous registration of TORCH_LIBRARY was registered at /dev/null:1953; latest registration was registered at /dev/null:1953

Also, I got this error from different environments, one with torch 2.3.1 and another one with torch 2.4.0.

Why does reloading torch trigger this error? And what could be done to successfully reload torch?

Originally, to reload torch, I tried:

import torch
import sys
sys.modules.pop('torch')
import torch

but I got another error:

---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-9-cfb33825cf77> in <cell line: 3>()
      1 # import sys
      2 # sys.modules.pop('torch')
----> 3 import torch

/usr/local/lib/python3.10/dist-packages/torch/__init__.py in <module>
    560     raise  # If __file__ is not None the cause is unknown, so just re-raise.
    561 
--> 562 for name in dir(_C):
    563     if name[0] != '_' and not name.endswith('Base'):
    564         __all__.append(name)

NameError: name '_C' is not defined