MemBoost: A Memory-Boosted Framework for Cost-Aware LLM Inference
This addresses cost reduction for LLM inference in interactive settings, but it is incremental as it builds on retrieval-augmented generation with added features like memory growth and routing.
The paper tackles the high inference cost of LLMs in real-world services by proposing MemBoost, a memory-boosted framework that reuses answers and retrieves information for cheap inference, reducing expensive large-model invocations while maintaining answer quality comparable to a strong model baseline.
Large Language Models (LLMs) deliver strong performance but incur high inference cost in real-world services, especially under workloads with repeated or near-duplicate queries across users and sessions. In this work, we propose MemBoost, a memory-boosted LLM serving framework that enables a lightweight model to reuse previously generated answers and retrieve relevant supporting information for cheap inference, while selectively escalating difficult or uncertain queries to a stronger model. Unlike standard retrieval-augmented generation, which primarily grounds a single response, MemBoost is designed for interactive settings by supporting answer reuse, continual memory growth, and cost-aware routing. Experiments across multiple models under simulated workloads show that MemBoost substantially reduces expensive large-model invocations and overall inference cost, while maintaining high answer quality comparable to the strong model baseline.