Implement to_device() for python bindings, add unit tests

.github/workflows/ci.yml

@@ -175,7 +175,7 @@ jobs:
         name: Set up QEMU
 
       - name: Build wheels
-        uses: pypa/cibuildwheel@v2.15.0
+        uses: pypa/cibuildwheel@v2.16.5
         with:
           package-dir: python
           output-dir: python/wheelhouse

python/cpp/storage_view.cc

@@ -106,6 +106,11 @@ namespace ctranslate2 {
         .value("int32", DataType::INT32)
         ;
 
+      py::enum_<Device>(m, "Device")
+        .value("cpu", Device::CPU)
+        .value("cuda", Device::CUDA)
+        ;
+
       py::class_<StorageView>(
         m, "StorageView",
         R"pbdoc(
@@ -206,6 +211,24 @@ namespace ctranslate2 {
                    A new ``StorageView`` instance.
              )pbdoc")
 
+        .def("to_device",
+             [](const StorageView& view, Device device) {
+               ScopedDeviceSetter device_setter(view.device(), view.device_index());
+               StorageView converted = view.to(device);
+               synchronize_stream(view.device());
+               return converted;
+             },
+             py::arg("device"),
+             py::call_guard<py::gil_scoped_release>(),
+             R"pbdoc(
+                 Converts the storage to another device.
+
+                 Arguments:
+                   device: The device to copy the data to.
+
+                 Returns:
+                   A new ``StorageView`` instance.
+             )pbdoc")
         ;
     }
 

python/ctranslate2/__init__.py

@@ -23,6 +23,7 @@
         AsyncScoringResult,
         AsyncTranslationResult,
         DataType,
+        Device,
         Encoder,
         EncoderForwardOutput,
         ExecutionStats,

python/tests/test_storage_view.py

@@ -47,6 +47,17 @@ def test_storageview_cpu(dtype, name):
     assert test_utils.array_equal(x, y)
 
 
+def test_storageview_to_device():
+    x = np.ones(10, dtype="float32")
+    cx = ctranslate2.StorageView.from_array(x)
+
+    assert cx.device == "cpu"
+    assert cx.dtype == ctranslate2.DataType.float32
+
+    cpu_x = cx.to_device(ctranslate2.Device.cpu)
+    assert test_utils.array_equal(x, np.array(cpu_x))
+
+
 @test_utils.require_cuda
 def test_storageview_cuda():
     import torch
@@ -74,6 +85,21 @@ def test_storageview_cuda():
     _assert_same_array(s.__cuda_array_interface__, y.__cuda_array_interface__)
 
 
+@test_utils.require_cuda
+def test_storageview_cuda_to_device():
+    x = np.ones(10, dtype="float32")
+    # convert to cuda tensor
+    cuda_x = ctranslate2.StorageView.from_array(x).to_device(ctranslate2.Device.cuda)
+    assert cuda_x.device == "cuda"
+
+    # modify original tensor and convert back
+    x *= 2
+    cpu_x = np.array(cuda_x.to_device(ctranslate2.Device.cpu))
+
+    assert cpu_x.dtype == x.dtype
+    assert x.sum() == 2 * cpu_x.sum()
+
+
 def test_storageview_conversion():
     x = np.ones((2, 4), dtype=np.float32)
     s = ctranslate2.StorageView.from_array(x)