⏱ 7 min read How speculative decoding speeds up token generation without changing model output, and the questions inference interviews use to test whether you understand it. Read article