>>8667932
well it's not as bad as that retard made it out to be, judging from the demo on hf, it's definitely worth tinkering with
>>8667994
i don't think you should call that an assistant prompt but i guess it's done that way it was done to let the encoder llm have some context
if it's a transformer-based translation model, you usually encode something like
"Translate the following text from Spanish to English: <Spanish text>"
and then decode
"<English-translated text>"