diff --git a/docarray/typing/tensor/torch_tensor.py b/docarray/typing/tensor/torch_tensor.py
index a78781f6a9b..32d1cbff7be 100644
--- a/docarray/typing/tensor/torch_tensor.py
+++ b/docarray/typing/tensor/torch_tensor.py
@@ -271,3 +271,10 @@ def _docarray_from_ndarray(cls: Type[T], value: np.ndarray) -> T:
     def _docarray_to_ndarray(self) -> np.ndarray:
         """cast itself to a numpy array"""
         return self.detach().cpu().numpy()
+
+    def new_empty(self, *args, **kwargs):
+        """
+        This method enables the deepcopy of `TorchTensor` by returning another instance of this subclass.
+        If this function is not implemented, the deepcopy will throw an RuntimeError from Torch.
+        """
+        return self.__class__(*args, **kwargs)
diff --git a/tests/units/typing/tensor/test_torch_tensor.py b/tests/units/typing/tensor/test_torch_tensor.py
index 0f3c9882e2a..02605142113 100644
--- a/tests/units/typing/tensor/test_torch_tensor.py
+++ b/tests/units/typing/tensor/test_torch_tensor.py
@@ -187,6 +187,22 @@ class MMdoc(BaseDoc):
     assert not (doc.embedding == doc_copy.embedding).all()
 
 
+def test_deepcopy_tensor():
+    from docarray import BaseDoc
+
+    class MMdoc(BaseDoc):
+        embedding: TorchTensor
+
+    doc = MMdoc(embedding=torch.randn(32))
+    doc_copy = doc.copy(deep=True)
+
+    assert doc.embedding.data_ptr() != doc_copy.embedding.data_ptr()
+    assert (doc.embedding == doc_copy.embedding).all()
+
+    doc_copy.embedding = torch.randn(32)
+    assert not (doc.embedding == doc_copy.embedding).all()
+
+
 @pytest.mark.parametrize('requires_grad', [True, False])
 def test_json_serialization(requires_grad):
     orig_doc = MyDoc(tens=torch.rand(10, requires_grad=requires_grad))