Remix.run Logo
treyd an hour ago

My understanding is that tokenization is largely serial, so for a large initial prompt it can make up a large chunk of input processing time since after handing it off to the model inference it's (able to be) fully parallel across all tokens.