* Fixed onnx test failures when run on a cpu backend * Updated check_torch_conversion function to include output comparison