MixLLM: Dynamic Routing in Mixed Large Language Models

TL;DR

MixLLM is a dynamic routing system for large language models that optimizes query assignment to balance response quality, cost, and latency, adapting over time with continual learning.

Contribution

We introduce MixLLM, a novel system using contextual bandits and continual training to effectively route queries among mixed LLMs, addressing dynamic trade-offs and model set changes.

Findings

Achieves 97.25% of GPT-4's quality at 24.18% of the cost

Outperforms baseline routing methods in response quality and efficiency

Adapts to evolving query patterns and model sets over time

Abstract

Large Language Models (LLMs) exhibit potential artificial generic intelligence recently, however, their usage is costly with high response latency. Given mixed LLMs with their own strengths and weaknesses, LLM routing aims to identify the most suitable model for each query in the stream to maximize response quality and minimize cost and latency. However, the challenges involve: (1) dynamic trade-offs among quality, cost, and latency; (2) enabling continual learning in deployed systems; and (3) navigating a varying (e.g., new LLM addition or old LLM removal) set of LLM candidates over time. To bridge these gaps, we develop MixLLM, a dynamic contextual-bandit-based routing system for query-LLM assignment. Specifically, we first leverage query tags to enhance query embeddings for the routing task. Next, we design lightweight prediction models to estimate the response qualities and costs of queries over LLMs. We then devise a meta-decision maker to choose the query-LLM assignments to best tradeoff response quality, cost, and latency. Finally, the system benefits from continual training, allowing it to adapt to evolving queries and user feedback over time. Our extensive experiments show that MixLLM achieves the best trade-offs in response quality, cost, and latency (97.25% of GPT-4's quality at 24.18% of the cost under the time constraint).