6104 remove deprecated tensor.storage usage (#6105)

Fixes #6104

### Description

workaround adapted from
https://github.com/pytorch/pytorch/blob/44dac51/torch/_tensor.py#L231-L233

### Types of changes
<!--- Put an `x` in all the boxes that apply, and remove the not
applicable items -->
- [x] Non-breaking change (fix or new feature that would not break
existing functionality).
- [ ] Breaking change (fix or new feature that would cause existing
functionality to change).
- [ ] New tests added to cover the changes.
- [ ] Integration tests passed locally by running `./runtests.sh -f -u
--net --coverage`.
- [x] Quick tests passed locally by running `./runtests.sh --quick
--unittests --disttests`.
- [x] In-line docstrings updated.
- [ ] Documentation updated, tested `make html` command in the `docs/`
folder.

Signed-off-by: Wenqi Li <[email protected]>

monai/data/__init__.py

@@ -122,18 +122,31 @@
     from multiprocessing.reduction import ForkingPickler
 
     def _rebuild_meta(cls, storage, dtype, metadata):
-        storage_offset, size, stride, meta_dict = metadata
+        storage_offset, size, stride, requires_grad, meta_dict = metadata
+        storage = storage._untyped_storage if hasattr(storage, "_untyped_storage") else storage
         t = cls([], dtype=dtype, device=storage.device)
-        t.set_(storage._untyped() if hasattr(storage, "_untyped") else storage, storage_offset, size, stride)
+        t.set_(storage, storage_offset, size, stride)
+        t.requires_grad = requires_grad
         t.__dict__ = meta_dict
         return t
 
     def reduce_meta_tensor(meta_tensor):
-        storage = meta_tensor.untyped_storage() if hasattr(meta_tensor, "untyped_storage") else meta_tensor.storage()
+        if hasattr(meta_tensor, "untyped_storage"):
+            storage = meta_tensor.untyped_storage()
+        elif hasattr(meta_tensor, "_typed_storage"):  # gh pytorch 44dac51/torch/_tensor.py#L231-L233
+            storage = meta_tensor._typed_storage()
+        else:
+            storage = meta_tensor.storage()
         dtype = meta_tensor.dtype
-        if storage.is_cuda:
+        if meta_tensor.is_cuda:
             raise NotImplementedError("sharing CUDA metatensor across processes not implemented")
-        metadata = (meta_tensor.storage_offset(), meta_tensor.size(), meta_tensor.stride(), meta_tensor.__dict__)
+        metadata = (
+            meta_tensor.storage_offset(),
+            meta_tensor.size(),
+            meta_tensor.stride(),
+            meta_tensor.requires_grad,
+            meta_tensor.__dict__,
+        )
         return _rebuild_meta, (type(meta_tensor), storage, dtype, metadata)
 
     ForkingPickler.register(MetaTensor, reduce_meta_tensor)

tests/test_meta_tensor.py

@@ -247,7 +247,7 @@ def test_torchscript(self, device):
                     "your pytorch version if this is important to you."
                 )
                 im_conv = im_conv.as_tensor()
-            self.check(out, im_conv, ids=False)
+        self.check(out, im_conv, ids=False)
 
     def test_pickling(self):
         m, _ = self.get_im()
@@ -258,7 +258,7 @@ def test_pickling(self):
             if not isinstance(m2, MetaTensor) and not pytorch_after(1, 8, 1):
                 warnings.warn("Old version of pytorch. pickling converts `MetaTensor` to `torch.Tensor`.")
                 m = m.as_tensor()
-            self.check(m2, m, ids=False)
+        self.check(m2, m, ids=False)
 
     @skip_if_no_cuda
     def test_amp(self):