Efficient Multi-Model Orchestration for Self-Hosted Large Language Models

TL;DR

This paper presents Pick and Spin, a scalable, cost-effective framework for orchestrating self-hosted large language models using Kubernetes, adaptive routing, and workload automation to improve performance and efficiency.

Contribution

It introduces a practical, hybrid routing and automation system for self-hosted LLMs, enhancing scalability, cost-efficiency, and reliability over traditional static deployment methods.

Findings

Up to 21.6% higher success rates

30% lower latency

33% lower GPU cost per query

Abstract

Self-hosting large language models (LLMs) is increasingly appealing for organizations seeking privacy, cost control, and customization. Yet deploying and maintaining in-house models poses challenges in GPU utilization, workload routing, and reliability. We introduce Pick and Spin, a practical framework that makes self-hosted LLM orchestration scalable and economical. Built on Kubernetes, it integrates a unified Helm-based deployment system, adaptive scale-to-zero automation, and a hybrid routing module that balances cost, latency, and accuracy using both keyword heuristics and a lightweight DistilBERT classifier. We evaluate four models, Llama-3 (90B), Gemma-3 (27B), Qwen-3 (235B), and DeepSeek-R1 (685B) across eight public benchmark datasets, with five inference strategies, and two routing variants encompassing 31,019 prompts and 163,720 inference runs. Pick and Spin achieves up to 21.6% higher success rates, 30% lower latency, and 33% lower GPU cost per query compared with static deployments of the same models.