fix aten::empty_like dtype (openvinotoolkit#22584)

src/frontends/pytorch/src/op/full.cpp

@@ -235,7 +235,7 @@ OutputVector translate_empty_like(const NodeContext& context) {
         if (!context.input_is_none(dtype_id)) {
             empty = base_translate_full_with_convert(context, sizes, value, dtype_id);
         } else {
-            empty = base_translate_full(context, sizes, value);
+            empty = base_translate_full_with_convertlike(context, sizes, value, input);
         }
     } else if (context.get_input_size() == 4) {
         auto out = context.input_is_none(3) ? input : context.get_input(3);

tests/layer_tests/pytorch_tests/test_empty.py

@@ -55,11 +55,11 @@ def _prepare_input(self, shape, dtype=np.float32, out=False):
             return (np.random.randn(*shape).astype(dtype if dtype is not None else np.float32),)
         return (np.random.randn(*shape), np.ones(shape, dtype=(dtype if dtype is not None else np.float32)))
 
-    def create_model(self, dtype, out):
+    def create_model(self, dtype, out, no_expose_dtype=False):
 
         class aten_empty_like(torch.nn.Module):
 
-            def __init__(self, dtype=None, out=False):
+            def __init__(self, dtype=None, out=False, no_expose_dtype=False):
                 dtype_map = {
                     "float32": torch.float32,
                     "float64": torch.float64,
@@ -72,6 +72,8 @@ def __init__(self, dtype=None, out=False):
                 self.dtype = dtype_map.get(dtype, None)
                 if out:
                     self.forward = self.forward_out
+                if no_expose_dtype:
+                    self.forward = self.forward_input_dtype
 
             def forward(self, input_tensor):
                 empty = torch.empty_like(input_tensor, dtype=self.dtype)
@@ -80,6 +82,14 @@ def forward(self, input_tensor):
                 # produce sporadic errors if nan would be in empty.
                 return torch.zeros_like(empty)
 
+            def forward_input_dtype(self, input_tensor):
+                # We don't want to compare values, just shape and type,
+                # so we call zeros_like on data. Multiplying by zero would
+                # produce sporadic errors if nan would be in empty.
+                input_tensor.to(self.dtype)
+                empty = torch.empty_like(input_tensor)
+                return torch.zeros_like(empty)
+
             def forward_out(self, input_tensor, out_tensor):
                 torch.empty_like(input_tensor, out=out_tensor)
                 # We don't want to compare values, just shape and type,
@@ -89,17 +99,25 @@ def forward_out(self, input_tensor, out_tensor):
 
         ref_net = None
 
-        return aten_empty_like(dtype, out), ref_net, "aten::empty_like"
+        return aten_empty_like(dtype, out, no_expose_dtype), ref_net, "aten::empty_like"
 
     @pytest.mark.parametrize('dtype', (None, "float32", "float64", "int64", "int32", "uint8", "int8"))
     @pytest.mark.parametrize("input_shape", [[2,], [1, 10], [10, 5, 2]])
     @pytest.mark.parametrize("out", [True, False])
     @pytest.mark.nightly
     @pytest.mark.precommit
-    def test_empty(self, ie_device, precision, ir_version, dtype, input_shape, out):
+    def test_empty_like(self, ie_device, precision, ir_version, dtype, input_shape, out):
         self._test(*self.create_model(dtype, out), ie_device, precision, ir_version, 
                    kwargs_to_prepare_input={"shape": input_shape, "out": out, "dtype": dtype})
 
+    @pytest.mark.parametrize('dtype', (None, "float32", "float64", "int64", "int32", "uint8", "int8"))
+    @pytest.mark.parametrize("input_shape", [[2,], [1, 10], [10, 5, 2]])
+    @pytest.mark.nightly
+    @pytest.mark.precommit
+    def test_empty_like_no_dtype(self, ie_device, precision, ir_version, dtype, input_shape):
+        self._test(*self.create_model(dtype, out=False, no_expose_dtype=True), ie_device, precision, ir_version, 
+                   kwargs_to_prepare_input={"shape": input_shape, "out": False, "dtype": dtype})
+
 
 class TestEmptyBoolean(PytorchLayerTest):