Add new Tensor ctor

src/bindings/python/src/pyopenvino/core/common.cpp

@@ -490,6 +490,10 @@ ov::Tensor create_shared(py::array& array) {
     OPENVINO_THROW("SHARED MEMORY MODE FOR THIS TENSOR IS NOT APPLICABLE! Passed numpy array must be C contiguous.");
 }
 
+ov::Tensor tensor_from_pointer(int64_t data_ptr, const ov::Shape& shape, const ov::element::Type& type) {
+    return ov::Tensor(type, shape, (void*)(data_ptr), {});
+}
+
 ov::Tensor tensor_from_pointer(py::array& array, const ov::Shape& shape, const ov::element::Type& type) {
     if (type_helpers::get_ov_type(array) == ov::element::string) {
         OPENVINO_THROW("SHARED MEMORY MODE FOR THIS TENSOR IS NOT APPLICABLE! String types can be only copied.");

src/bindings/python/src/pyopenvino/core/common.hpp

@@ -157,6 +157,8 @@ T object_from_data(D& data, bool shared_memory) {
     return create_copied<T>(data);
 }
 
+ov::Tensor tensor_from_pointer(int64_t data_ptr, const ov::Shape& shape, const ov::element::Type& ov_type);
+
 ov::Tensor tensor_from_pointer(py::array& array, const ov::Shape& shape, const ov::element::Type& ov_type);
 
 ov::Tensor tensor_from_pointer(py::array& array, const ov::Output<const ov::Node>& port);

src/bindings/python/src/pyopenvino/core/tensor.cpp

@@ -37,6 +37,38 @@ void regclass_Tensor(py::module m) {
                 :type shared_memory: bool
             )");
 
+    cls.def(py::init([](int64_t data_ptr, const ov::Shape& shape, const ov::element::Type& ov_type) {
+                return Common::tensor_from_pointer(data_ptr, shape, ov_type);
+            }),
+            py::arg("data_ptr"),
+            py::arg("shape"),
+            py::arg("type") = ov::element::undefined,
+            R"(
+                Tensor's special constructor .
+                Represents an array in the memory with given shape and element type.
+                It's recommended to use this constructor when initializing Tensor
+                with data stored in non-RAM memory (for example GPU memory). It creates
+                an OpenVINO Tensor without the need for additional data copies.
+                :param data_ptr: Pointer to a C_CONTIGUOUS array which will be wrapped in
+                                 openvino.runtime.Tensor with given parameters (shape
+                                 and element_type). Array's memory is being shared with
+                                 a host, that means the responsibility of keeping host memory is
+                                 on the side of a user. Any action performed on the host
+                                 memory will be reflected on this Tensor's memory!
+                :type data_ptr: int
+                :param shape: Shape of the new tensor.
+                :type shape: openvino.runtime.Shape
+                :param type: Element type
+                :type type: openvino.runtime.Type
+                :Example:
+                .. code-block:: python
+                    from openvino import Tensor, Shape, Type
+                    import torch
+                    arr = torch.rand(128).to(torch.device("xpu"))
+                    data_ptr = arr.detach().data_ptr()
+                    t = Tensor(data_ptr, Shape(arr.shape), Type.f32)
+            )");
+
     cls.def(py::init([](py::array& array, const ov::Shape& shape, const ov::element::Type& ov_type) {
                 return Common::tensor_from_pointer(array, shape, ov_type);
             }),

src/bindings/python/tests/test_runtime/test_tensor.py

@@ -617,3 +617,11 @@ def get_tensor():
 
     tensor = get_tensor()
     assert np.allclose(tensor.data[0][0][0:3], [0, 0, 1])
+
+
+def test_init_from_data_ptr():
+    arr = np.ones((8, 16, 300), dtype=np.float32)
+    pointer, _ = arr.__array_interface__['data']
+    tensor = ov.Tensor(pointer, ov.Shape(arr.shape), ov.Type.f32)
+    arr[0][0][0:2] = 0
+    assert np.allclose(tensor.data[0][0][0:3], [0, 0, 1])