为什么使用GPU比CPU要慢呢？

使用的示例进行测试：

```
from reazonspeech.k2.asr import load_model, transcribe, audio_from_path
import time


audio = audio_from_path("ja_test.wav")
model = load_model("cuda", "int8", "ja")
start_time = time.monotonic()
ret = transcribe(model, audio)
print(ret.text)
print(f"Decoding took {time.monotonic() - start_time:.2f} seconds")```


在4090和5090上都试过，结果是一样的。1个6s的音频文件，CPU耗时0.23s, GPU耗时0.66s