3 ms·Co-Designing AI Models Using Speculative Decoding for Faster LLM Inference2 points by buildbot 2mo ago