I can not get expected result following the quick start demo

I did it as link https://opennmt.net/CTranslate2/quickstart.html

c++ code as below:

`int main()
{
//system("chcp 850");

const std::string model_path("D:\\code\\CTranslate2\\models\\opus-mt-en-de\\");
const ctranslate2::models::ModelLoader model_loader(model_path);

ctranslate2::Translator translator(model_loader);

const std::vector<std::vector<std::string>> batch = { {"▁Hello", "▁World", "!", "</s>"} };
const auto translation = translator.translate_batch(batch);

for (const auto& token : translation[0].output())
    std::cout << token << ' ';
std::cout << std::endl;

return 0;

}`

but the output is:

鈻? 鈻丯ein , 鈻乶ein .

instead of '▁Hall o ▁Welt !' that in document.

my model is generated from following command:

huggingface-cli download --resume-download --local-dir-use-symlinks False Helsinki-NLP/opus-mt-en-de
cd D:\code\CTranslate2\models
ct2-transformers-converter --model Helsinki-NLP/opus-mt-en-de --quantization int8 --output_dir opus-mt-en-de

I need help.
sorry for my poor English.

Start with the C++ quick-start snippet and the Hugging Face download and ct2-transformers-converter commands shown in the issue. Compare the generated model's translation output with the quick-start expectation, including the displayed token encoding. Done means the same model and C++ example produce the documented output or the cause of the discrepancy is documented.