Search papers, labs, and topics across Lattice.
Coral is introduced as a multi-LLM serving system that adaptively optimizes resource allocation and serving strategies across heterogeneous cloud GPUs. It employs a lossless two-stage decomposition to maintain joint optimality while drastically reducing online solve time for resource allocation. Experiments across 6 models and 20 GPU configurations demonstrate that Coral achieves up to 2.79x cost reduction and 2.39x higher goodput compared to baseline methods, especially under resource constraints.
Save up to 2.79x on LLM serving costs by intelligently distributing models across a diverse fleet of cloud GPUs.
The usage of large language models (LLMs) has grown increasingly fragmented, with no single model dominating. Meanwhile, cloud providers offer a wide range of mid-tier and older-generation GPUs that enjoy better availability and deliver comparable performance per dollar to top-tier hardware. To efficiently harness these heterogeneous resources for serving multiple LLMs concurrently, we introduce Coral, an adaptive heterogeneity-aware multi-LLM serving system. The key idea behind Coral is to jointly optimize resource allocation and the serving strategy of each model replica across all models. To keep pace with shifting throughput demand and resource availability, Coral applies a lossless two-stage decomposition that preserves joint optimality while cutting online solve time from hours to tens of seconds. Our evaluation across 6 models and 20 GPU configurations shows that Coral reduces serving cost by up to 2.79$\times$ over the best baseline, and delivers up to 2.39$\times$ higher goodput under scarce resource availability.