LiveServe: Interaction-Aware Serving for Real-Time Omni-Modal LLMs
For developers and operators of real-time omni-modal LLM services, this work addresses the inefficiency of throughput-oriented scheduling and LRU-based KV eviction in interactive speech-centric applications.
LiveServe introduces an interaction-aware serving system for real-time omni-modal LLMs that leverages playback progress, speech activity, and barge-in events to improve scheduling and KV cache management. It reduces P90 audio time-to-first-playback by up to 2.21× and increases completed-request throughput by up to 1.56× compared to vLLM-Omni.
Realtime omni-modal LMs support speech-centric conversations where users stream inputs, hear generated audio, and interrupt freely. Existing Omni-LM serving systems still rely on throughput-oriented LLM scheduling and LRU KV offloading. These policies ignore audio playback and multi-turn reuse: they may generate tokens far beyond what users hear, wasting work after barge-in, and evict KV state needed in the next turn. LiveServe is an interaction-aware serving system for realtime Omni-LM interaction. It exposes playback progress, speech activity, and barge-in events to the serving pipeline. The scheduler prioritizes first-audio and near-underrun sessions while limiting generation beyond the playback frontier. The KV manager uses next-use-aware eviction and preloads likely-needed KV during user speech to hide reload latency. On vLLM-Omni, LiveServe improves realtime serving across two Omni-LMs and mixed workloads. It lowers P90 audio TTFP by $1.55\times$ on average and up to $2.21\times$, while improving completed-request throughput by $1.15\times$ on average and up to $1.56\times$, and moves most KV reload work off the next-turn critical path.