>>107140935
are you using moonshot as the chat completion source?
check your console, when you see the prefill message, does it have partial: true in it?
e.g
messages: [
{ role: 'user', content: 'Hi Kimi.' },
{
role: 'assistant',
content: "<think>\nOh no, I'm being prefilled!",
partial: true
}
],