In this vLLM Office Hours session, we kicked off with the latest updates across the vLLM project, shared by core committer Michael Goin. The session then transitioned into a special topic focused on Speculators, a new unified library for building, storing, and experimenting with speculative decoding algorithms in vLLM. Fynn Schmitt-Ulms, Helen Zhao, and Rahul Tuli from Red Hat introduced the Speculators project, covering the motivation behind the library, how it fits into the vLLM ecosystem, and how developers can use it to standardize and experiment with speculative decoding techniques. Timestamps: 00:00 – Welcome and what's new in the vLLM community over the past week 11:02 – Deep dive into to Speculators 48:34 – Q&A and Discussion Session slides: https://docs.google.com/presentation/d/1s4eAb7v-rdZt8smyULBJWGXjJXrgFTZWnwqYa2-h1l4/ Speculators GitHub repository: https://github.com/vllm-project/speculators If you want to join future sessions live on Google Meet and take part in the discussion, you can register for upcoming vLLM Office Hours here: https://red.ht/office-hours

 563      16