[CPU] Fix ScaledDotProductAttension accuracy problem (#21217)

tests/model_hub_tests/torch_tests/test_timm.py

@@ -75,7 +75,8 @@ class TestTimmConvertModel(TestConvertModel):
 
     @pytest.mark.parametrize("name", ["mobilevitv2_050.cvnets_in1k",
                                       "poolformerv2_s12.sail_in1k",
-                                      "vit_base_patch8_224.augreg_in21k"])
+                                      "vit_base_patch8_224.augreg_in21k",
+                                      "beit_base_patch16_224.in22k_ft_in22k"])
     @pytest.mark.precommit
     def test_convert_model_precommit(self, name, ie_device):
         self.run(name, None, ie_device)