The ever-growing sizes of large language models (LLMs) introduce significant infrastructure challenges due to their immense memory capacity demands. While the de facto approach has been to deploy multiple high-end GPUs, each with a limited memory capacity, the prohibitive cost of such systems has become a major barrier to the widespread deployment of frontier LLMs. As a result, CPU-based inference has become an appealing and cost-efficient alternative, since a CPU can offer an order of magnitude larger memory capacity at a fraction of the cost while providing competitive throughput for matrixvector multiplication with the latest Advanced Matrix Extensions (AMX). It not only broadens accessibility for users without multiGPU setups but also enables hyperscalers to leverage underutilized CPU servers to accommodate temporarily surging inference demand. Nevertheless, even CPU's large memory capacity has become insufficient to serve LLMs with hundreds of billions of parameters. Under the memory capacity constraint, we may offload parameters to storage devices and fetch them on demand, but doing so significantly degrades inference performance due to the high latency and low bandwidth of storage devices. To address this challenge, we propose LILO, an LLM inference framework that leverages In-memory Analytics Accelerator (IAA) in the latest Intel CPUs, to accelerate inference under memory capacity constraints. By storing model parameters in a compressed format and decompressing them on demand using IAA, LILO enables significantly reduced storage access during inference under memory capacity constraints while preserving the model accuracy and behavior. LILO orchestrates the concurrent execution of on-chip accelerators, i.e., IAA, Advanced Vector Extensions (AVX), and AMX, to facilitate high-throughput decompression alongside inference computation. Furthermore, LILO implements selective compression, a Mixture-of-Expert (MoE)-aware optimization that reduces the decompression overhead by up to 1.9×. We demonstrate that LILO reduces inference latency by up to 4.9× and 4.3× for Llama3-405B and DeepSeekR1, respectively, under memory capacity constraints compared to the baseline inference solely relying on storage-offloading without compression.

Combinatorial auctions (CAs) offer the flexibility for bidders to articulate complex preferences when competing for multiple assets. However, the behavior of bidders under different payment rules is often unclear. Our research explores the relationship between core constraints and several core-selecting payment rules. Specifically, we examine the natural and desirable property of payment rules of being non-decreasing, which ensures that bidding higher does not lead to lower payments. Earlier studies revealed that the VCG-nearest payment method – a commonly employed payment rule – fails to adhere to this principle even for single-minded CAs. We establish that when a single effective core constraint exists, the payment maintains the non-decreasing property in single-minded CAs. To identify auctions where such a constraint is present, we introduce a novel framework using conflict graphs to represent single-minded CAs and establish sufficient conditions for the existence of single effective core constraints. We proceed with an analysis of the implications on bidder behavior, demonstrating that there is no overbidding in any Nash equilibrium when considering non-decreasing core-selecting payment rules. Our study concludes by establishing the non-decreasing nature of two additional payment rules, namely the proxy and proportional payment rules, for single-minded CAs.