.onnx Model file is int64_t.Can it support int16_t ？

[High-calcium-tiger]

I noticed that the CPU usage is very high when running piper tts on my Linux device.
I want change int64_t to int16_t to reduce CPU usage.
code：
std::vectorOrt::Value inputTensors;
std::vector<int64_t> phonemeIdsShape{1, (int64_t)phonemeIds.size()};
inputTensors.push_back(Ort::Value::CreateTensor<int64_t>(
memoryInfo, phonemeIds.data(), phonemeIds.size(), phonemeIdsShape.data(),
phonemeIdsShape.size()));

I change CreateTensor<int64_t> to CreateTensor<int16_t>，but an error occurred:Expectation int64_t.
I guess .onnx model file is int64_t,
I want to know Can it support int16_t ？