The very first run needs to wait for the machine translation model to be downloaded so it won't be quick. For the subsequent runs, the model will be reused as a cache.