knowing the size of download

[jimmywarting]

there might be scenarios where client do NOT want to download a model if it will be too large to download over mobile network or dose not want to take up that much space.
so knowing it in advance would be useful.

i also tried loading my own language code and it fails

it would be a lot easier to just be able to enumerate all the supported language model, and how large they are and if they are already downloaded or not and present them in a <select> for the user to be able to choose between instead of playing a guessing game of what is available or not.

speechSynthesis have something similar already: speechSynthesis.getVoices()

[tomayac]

We on purpose do not expose model sizes to avoid people depending on it (as in: ah, 1.234GB, this must be Gemini Nano v456). The user agent should be smart enough to decide if it's adequate to download the model given the current situation. We have webmachinelearning/prompt-api#158 open as an umbrella Issue for making model features (like here the supported languages) queryable.