4 ms·Speculative Speculative Decoding: Really, Really Fast LLM Inference1 points by fizzbuzz07 7mo ago