Marconi: Prefix Caching for the Era of Hybrid LLMs
This work addresses a specific bottleneck in serving hybrid LLMs for long-context applications, offering incremental improvements in caching efficiency.
The paper tackles the problem of inefficient prefix caching in hybrid LLMs, which combine attention and recurrent layers, by introducing Marconi, a system with novel admission and eviction policies that achieves up to 34.4× higher token hit rates and 71.1% lower TTFT compared to state-of-the-art systems.
Hybrid models that combine the language modeling capabilities of Attention layers with the efficiency of Recurrent layers (e.g., State Space Models) have gained traction in practically supporting long contexts in Large Language Model serving. Yet, the unique properties of these models complicate the usage of complementary efficiency optimizations such as prefix caching that skip redundant computations across requests. Most notably, their use of in-place state updates for recurrent layers precludes rolling back cache entries for partial sequence overlaps, and instead mandates only exact-match cache hits; the effect is a deluge of (large) cache entries per sequence, most of which yield minimal reuse opportunities. We present Marconi, the first system that supports efficient prefix caching with Hybrid LLMs. Key to Marconi are its novel admission and eviction policies that more judiciously assess potential cache entries based not only on recency, but also on (1) forecasts of their reuse likelihood across a taxonomy of different hit scenarios, and (2) the compute savings that hits deliver relative to memory footprints. Across diverse workloads and Hybrid models, Marconi achieves up to 34.4$\times$ higher token hit rates (71.1% or 617 ms lower TTFT) compared to state-of-the-art prefix caching systems.